Assessing climate change risks under uncertain conditions

Assessing climate change risks under uncertain conditions

Antonello Provenzale, Elisa Palazzi (1)

1ISAC-CNR, Torino

Climate and environmental change is expected to affect hydrometeorological haz-ard and ecosystem functioning, with possible threats to human societies due to in-creased probability of extreme events and loss of ecosystem services. In mountain regions, the environmental response could be even larger. For this reason, it is im-portant to obtain estimates of the expected modifications in natural hazards asso-ciated with climate and environmental change, to develop appropriate adaptation and risk mitigation strategies. This goal, however, is made difficult by the scale mismatch between climate model projections and land surface response, which re-quires the use of appropriate climate downscaling procedures. To complicate the picture, one should also cope with the chain of uncertainties which affect climate and risk projections, from the wide range of global climate model estimates for the water cycle variables, to the uncertainties in regional climate response, to the un-certainties in the hydrological and/or ecosystem models themselves. Precipitation data used to validate the models, on the other hand, are also affected by severe un-certainties, especially in mountain regions. This leads to the general problem of assessing natural hazards for different climate and environmental change scenarios under uncertain conditions. Keywords: Climate Change, Risk, Uncertainty assessment

1. Introduction

Global warming should not be intended as a slow, continuous and homogene-ous temperature rise. Temporally, the warming of the last eighty years has mani-fested itself in a sequence of steps, alternating periods of rapid change with times of slowly changing temperatures (as in the last ten years). Spatially, the tempera-ture increase is widely varying, with geographical areas where the warming has been up to three times the global mean (as in the Arctic and in several mountain chains such as the Alps), and other regions where warming has been more modest (IPCC 2013).

Even more importantly, the effect of global warming is not solely a temperature increase. The largest energy of the atmospheric motions and the potentially larger amount of moisture in the atmosphere can lead to an intensification of the hydro-logical cycle, with more intense precipitation events and more prolonged dry peri-

2

ods. Evidence of this behaviour is now available, and these effects are expected to become even more relevant in coming decades (Giorgi et al. 2011).

In turn, stronger intensity of the hydrological cycle is associated with potential modifications in the hydro-meteorological risk. For this reason, assessing the im-pact of climate change on the statistics of precipitation extremes, and the associat-ed surface effects such as flooding and landslides, is a much needed step in order to develop adequate risk mitigation and adaptations strategies. Of course, science can help in suggesting the proper avenues, but it is then in the hands of administra-tors and politicians to assure that the proper actions are taken.

2. Climate downscaling and surface processes

To estimate the impact of future climate change on surface geo-hydrological and ecosystem processes, we have to face a difficult problem: information on cli-mate projections is usually provided on relatively large spatial scales (50-100 km for global climate models and 10-50 km for most regional climate models), while the hydrological response takes place at much smaller spatial scales, particularly in mountain areas and small coastal basins. Thus, a climate downscaling proce-dure is needed, especially for highly intermittent (but crucial) fields such as pre-cipitation.

In recent years, various ways to bridge the scale gap between climate change scenarios and the small scales needed for impact studies have been proposed. One option is provided by statistical and/or stochastic downscaling methods. Statistical downscaling maps large-scale deterministic predictors to variables at small scales (Maraun et al. 2010), to produce realizations of the expected small-scale climate variability. Stochastic rainfall downscaling (Rebora et al. 2006, D'Onofrio et al. 2014) aims at generating synthetic spatial-temporal precipitation fields whose sta-tistical properties are consistent with the small-scale statistics of observed precipi-tation, based only on the knowledge of the large-scale precipitation field. Stochas-tic downscaling also holds the potential of estimating uncertainties in rainfall scenarios, by generating ensembles of small-scale precipitation fields which can be compared with measured data (Brussolo et al. 2008).

Clearly, stochastic downscaling is not a substitute for physically-based models, and it is just a way to introduce realistic rainfall variability on the scales which are not resolved by physical models. Dynamical downscaling, on the other hand, at-tempts at nesting high-resolution, non-hydrostatic convective models in larger-scale global or regional climate models, producing a small-scale zoom of the rele-vant dynamical fields. Of course, such approach is much more powerful and inter-esting, but the computational burden (in terms of CPU and storage) of these simu-lations often limits the possibility of having large ensembles of realizations or long time slices, thus reducing the ability to estimate climate statistics.

3

Thus, at the time of writing, the standard "climate impact" chain is to start with large-scale global climate models (or reanalysis products if one wants to verify current conditions), go through a nested (hydrostatic or non-hydrostatic) regional climate model, then through a statistical/stochastic downscaling procedure, to fi-nally drive eco-hydrological and geo-hydrological models for risk assessment. Such chain is illustrated in Figure 1.

Figure 1 – The climate downscaling chain for precipitation.

Global&climate&model&

Regional&climate&model&

Sta1s1cal/stochas1c&downscaling&&

Impact&on&eco8hydrological&processes&

4

3. The chain of uncertainties

Parallel to the downscaling chain, we should consider another important ele-ment of the story: the chain of uncertainties propagating from the global climate models through the downscaling procedures, to the final risk estimate. This aspect is far less studied, and it can potentially affect our ability to assess the impact of climate change on surface processes.

First, we have to take into account the uncertainties arising from global climate simulations. These are generated by the uncertainty in the emission scenario which will be followed - and for this reason we should always consider different scenari-os, such as the currently utilized RCP4.5, RCP8.5 and so on - but also by the natu-ral variability of climate and of its numerical representations for a given scenario (see for example Provenzale 2014).

Figure 2 shows the time series of surface mean winter temperature, spatially-averaged over the Hindu-Kush Karakoram area in Asia, as simulated by an en-semble of global climate models from the CMIP5 initiative (http://cmip-pcmdi.llnl.gov/cmip5/). The black line represents the average over all model re-sults for the historical period, the blue and red lines represent the average over all models for the RCP4.5 and RCP8.5 future scenarios respectively, and the pink and green lines represent the data from CRU and GHCN respectively. The pale grey lines represent the outputs of the individual models, and their spread indicates the range of model responses.

Figure 2 – Average winter temperature, as a function of time, for the area of the Hindu-Kush Karako-ram (HKK) in Asia, as simulated by a suite of global climate models from the CMIP5 initiative. The meaning of the different lines is discussed in the text.

5

From this figure, some interesting facts emerge: (1) there is not much agree-ment between the two observational datasets (even though they do measure exact-ly the same quantity: surface temperature for CRU and 2m air temperature for GHCN); (2) most models predict a cooler climate than observed today (the so-called "cold bias"), and (3) the average winter temperature displays a difference of up to ten degree between the different models. Clearly, both issues are serious, and require further work on climate models - something that most workers in this field are aware of. On the other hand, even the validation of model results on measured data is difficult, especially in the case of precipitation, since in certain regions such as the Karakoram-Himalayan chain the different datasets provide rather dif-ferent views and a definite "ground truth" is not available (Palazzi et al. 2013).

But if we are interested in driving a downscaling/impact chain with such mod-els, what should we do? Often, what is done is to use some form of "bias correc-tion": we normalize the model output to the currently observed values (of tem-perature or precipitation), and then we simply look at the anomaly, that is, at the difference between the model-generated values for the future and those for today. This procedure is simple, and widely used, but it is not devoid of danger. For ex-ample, a cold bias affects the saturation water vapour pressure in the atmosphere (where temperature enters nonlinearly and not as an anomaly), with possibly seri-ous consequences for the correct reproduction of the hydrological cycle. For this reason, we cannot take climate model outputs at face value and simply bias-correct them, but we should be aware of the spread in model behaviours and of its potential implications.

When nesting a regional model for a specific area in the output of a global cli-mate simulation, one drives the regional model by the boundary conditions pro-vided by the global model. But then, what is the effect of uncertainties in such boundary values and how such uncertainties propagate in the regional model? That is, what is the sensitivity of the regional model to changes in boundary condi-tions? Such issue has been explored for specific cases (see for example the Arctic case considered by Rinke and Dethloff 2000), and should not be forgotten when working with nested regional models. In addition, regional models themselves can add their own uncertainty and unpredictability, which is often difficult to quantify.

The statistical/stochastic downscaling adds its own uncertainties, mainly related to the fact that statistical relationships and stochastic realizations do not provide a single time series of temperature or precipitation in the location of interest, but ra-ther give an ensemble of possible time histories, all compatible with the con-straints imposed by the climate models at the larger scales. An interesting point is to determine how the uncertainty range introduced by the downscaling compares with that associated with the ensemble of climate models. The question is not sim-ple and must be carefully disentangled, see for example von Hardenberg at al. (2007) for a similar problem in a meteorological framework.

Finally, we have the uncertainty coming from the imperfect knowledge of the model for surface processes, such as the form and parameter values for a semi-distributed hydrological model, or the soil and vegetation parameters in an eco-

6

hydrological model. The relevance of such uncertainty varies with the model and the processes considered, and it could either be larger or smaller than the uncer-tainty propagated through the chain till this point. In such conditions, one should also consider the sensitivity of a given surface process model to variations in the driving factors. An example of the assessment of the sensitivity of a hydrological model to variations in the parameters of the stochastic downscaling procedure is discussed in Gabellani et al. (2007).

4. Final remarks Assessing the effects of climate change on surface geo-eco-hydrological pro-

cesses, and estimating the associated risks, is currently one of the most important challenges for the Geosciences. This challenge involves important conceptual as-pects, such as cross-scale interactions and downscaling methods, and it has crucial societal implications. However, several sources of uncertainty contribute to make this endeavour rather complex. In particular, we cannot provide useful estimates of the surface response and of the potential risks if we do not include a quantita-tive assessment of the uncertainty on the estimates. As seen above, the different sources of uncertainty all potentially contribute to affect the final result, and they should be carefully quantified and weighted. Most current impact assessments in-clude the use of different scenarios, and many take into account model variability by considering an ensemble of model behaviours. However, this is often not enough, as other uncertainty sources are at play. On the other hand, it can be diffi-cult from a practical point of view to build ensembles of ensembles of ensembles of realizations taking into account all possible uncertainties and their propagation across scales, and new, more refined, ingenious and synthetic approaches should be developed.

For these reasons, the assessment of climate change impacts on surface pro-cesses is not just an automated procedure based on the blind applications of mod-els driven by other models, but it is, in all respect, an intriguing, challenging and difficult scientific endeavour. It is, in addition, a topic having extremely important practical implications, as it is at the base of the development of knowledge-based risk mitigation and adaptation strategies. Such topic deserves the establishment of national and international programs devoted to provide risk assessments in an un-certain future.

7

[References]

IPCC (2013) - Fifth Assessment Report (AR5), Approved Summary for Policymak-ers. Intergovernmental Panel on Climate Change. http://www.ipcc.ch. Brussolo E., von Hardenberg, J., Ferraris, L., Rebora, N., Provenzale, A. 2008. Verification of Quantitative Precipitation Forecasts via Stochastic Downscaling. J. Hydrometeorology 9, 1084-1094. D'Onofrio D., Palazzi, E., von Hardenberg, J., Provenzale, A., Calmanti, S. 2014. Stochastic Rainfall Downscaling of Climate Models. J. Hydrometeorology, http://dx.doi.org/10.1175/JHM-D-13-096.1. Gabellani, S., Boni, G., Ferraris, L., von Hardenberg, J., Provenzale, A. 2007. Propagation of uncertainty from rainfall to runoff: A case study with a stochastic rainfall generator. Adv. Water Resources 30, 2061-2071.

Giorgi, F., Im, E.-S., Coppola, E., Diffenbaugh, N.S., Gao, X.J., Mariotti, L., Shi, Y., 2011. Higher Hydroclimatic Intensity with Global Warming. J. Climate 24, 5309-5324.

von Hardenberg, J., Ferraris, L., Rebora, N., Provenzale, A., 2007. Meteorological uncertainty and rainfall downscaling. Nonlinear Processes in Geophysics 14, 193-199.

Maraun, D., et al. 2010. Precipitation downscaling under climate change: Recent developments to bridge the gap between dynamical models and the end user. Rev. Geophys., 48, RG3003, doi:10.1029/2009RG000314.

Palazzi, E., von Hardenberg, J., Provenzale, A. 2013. Precipitation in the Hindu-Kush Karakoram Himalaya: Observations and future scenarios. J. Geophys. Res. Atmospheres 118, doi:10.1029/2012JD018697. Provenzale, A., 2014. Climate models. Rendiconti Lincei 25, 49-58.

Rebora, N., Ferraris, L., von Hardenberg, J., Provenzale, A. 2006. RainFARM: Rainfall Downscaling by a Filtered Autoregressive Model. J. Hydrometeorology 7, 724-738.

Rinke, A., Dethloff, K. 2000. On the sensitivity of a regional Arctic climate model to initial and boundary conditions. Climate Research 14, 101-113.