Context

In the short term (next 30 to 50 years), climate change will definitely have the greatest impact in tropical and subtropical regions with highly variable climates and little capacity to adapt to this variability, making them particularly vulnerable to climate hazards. Assessing this impact currently involves substantial uncertainty, particularly regarding subgrid variables such as precipitation, which is poorly simulated for the current period. The scale of precipitation data is insufficient to serve as input for impact models, and in several tropical regions, future changes in precipitation diverge depending on the climate model used. For crops that are primarily rainfed, as well as for river flow rates and floodplain evolution, local changes in precipitation (means as well as synoptic, seasonal and interannual variability) are decisive factors. 

Objectives

REGYNA's aim is two-fold: to identify the sources of uncertainty in order to better assess the short-term hydrological and agricultural impacts, and to study the societal consequences in three vulnerable regions - the Mediterranean region, West Africa and the La Plata and southern Amazon basins. The project's methodology will be general: it will focus on three regions that are very different, both in terms of climate variability (Sahel Africa has experienced a severe drought since the end of the 1960s, whereas rainfall has been increasing in tropical and subtropical South America since that time) and in terms of future evolution (the models agree that climate change will dry out the Mediterranean basin, whereas forecasts diverge for West Africa). The methodology will also be adaptable: the project will define a metric for ranking the various models based on their biases on the present and the future. This in turn could be used to assess other simulations outside the scope of REGYNA. 

Methodology

Phase 1: Links between the large scale and the local scale

The aim is to characterise the links between large-scare atmospheric circulation and local precipitation in order to statistically reproduce local daily rainfall, a key factor for climate impact studies.

1) Weather types

The first step will use the NCEP and ERA40 atmospheric reanalyses to identify daily atmospheric situations which, by their persistence, alternation and frequency, modulate climate variability in the regions studied. Once these weather types or regimes are established for each region, researchers will analyse their frequency, persistence and the transitions from one type to another. This will result in an exhaustive description of climate variability in these regions, to be compared with the data in the literature.

2) Links with rainfall

The project will also describe the links between the weather regimes and precipitation. For each of the three regions of interest, a rainfall database is available on the scale of the station covering at least 30 years. These data can be used to assess the effects of large-scale atmospheric events on local precipitation by establishing a transfer function between the weather type classification and the local rainfall characteristics (frequency, intensity, wet or dry periods, etc.). The statistical approach developed will characterise normal and extreme rain events, to optimally reproduce local precipitation in each of the three regions of interest, rather than maximising the method's performance in a single region. The method will also be assessed using the database for the current period: errors and uncertainties, ability to describe the frequency of daily rainfall values as well as their seasonal and interannual variability.

Part 2: Assessing uncertainties in the context of climate change

This part of the project will regionalise precipitation in the context of climate change. Researchers will set out to quantify the hierarchy of uncertainties associated with regionalisation and climate forecasts. Quantifying uncertainty is critical to climate impact studies and will be the guiding principle for this segment.

1) Uncertainties associated with climate forecasts

A two-step methodology will be used to assess uncertainties relating to differences between the models and to climate forecasts:
- For two greenhouse gas emission scenarios (SRESA2 and SRESB1), the biases and forecasts of IPCC models will be evaluated for each region of study. Researchers will study the possibility of combining the models and look for any links between the biases and forecasts. The results should quantitatively measure the ability of the models to reproduce the precipitation structures observed at different timescales and predict their evolution in response to climate change. The goal of this step is to rank the models both on how they represent the current climate and on how they change in the future.
- The models to be used for the impact studies will be selected.

2) Uncertainties associated with model performance

While the primary objective of regionalisation is to downscale the phenomena identified by the climate models, this reduction in scale will be insufficient unless accompanied by better representation of the means and variability of the current climate, including extreme events. For the selected models, improved simulation of precipitation for the present period will be demonstrated in two steps:
- Assess the ability of the models to reproduce the atmospheric variables necessary for regionalising precipitation and to define weather types;
- Apply the transfer function defined in Part 1 to data from the various models, and comparing the local rainfall thus obtained with observed local rainfall. Researchers expect better results than those obtained with current climate models (at comparable scales).

3) Uncertainties associated with the climate change signal

For each selected model, several simulations will be conducted (same scenario but slightly different initial conditions) to measure the model's internal variability (including decadal and multidecadal variability). This variability will be compared to the magnitude of the climate change signal in the near future (next 30 to 50 years). Researchers can then assess whether the model forecasts are associated with long-term climate change or with decadal or multidecadal variability.

4) Uncertainties associated with future changes in weather types

The statistical regionalisation method developed in Part 1 is based on current links between large-scale weather types and local precipitation. Given that these relationships are not guaranteed in a modified climate, is it possible to apply the regionalisation method to climate change scenarios, and if so, what uncertainties does this involve? This question will be approached in several different ways:
- The method's stability will be established by an assessment process conducted on three different regions. This makes it impossible to introduce characteristics that are too local in the transfer function, avoiding a potential source of instability in the context of climate change.
- In view of applying regionalisation to future climates, current atmospheric structures are assumed to remain unchanged while their frequency is assumed to evolve. This assumption will be tested on the selected models.
- Finally, as a way around changes in weather type, a very simple statistical method is proposed to generate future local fields (precipitation and/or temperatures) by modifying the current observed series in such a way that they reproduce the mean and variability changes of a climate change scenario. This type of basic method can be applied to each model and scenario selected in order to serve as a reference for analysing more sophisticated methods.

Part 3: Impact of climate change in vulnerable regions

1) Evaluating hydrological risks in South America

The objective is to assess the evolution of the river flow rates and floodplain characteristics of the southern Amazon basin (Madeira basin in Bolivia, south-west Brazil) and the La Plata basin. This will make it possible to evaluate the vulnerability and operational sustainability of several infrastructure projects (river ports between Rosario and Buenos Aires, hydropower stations on the Madeira and Amazon water routes), and to map out areas presenting risks for the people, livestock and grain crops around the cities (Buenos Aires) and the vast floodplains (Llanos de Moxos in Bolivia, Entre Rios, Corrientes, Sante Fe).

Step 1: Better understand the links between rainfall and river flow rates. This research will be conducted using information from databases that the Hybam environmental research observatory has built for several southern Amazon basins (of variable size and subject to very high seasonal and interannual variability). Modelling rainfall-flow rate in this way will make it possible to determine evolution scenarios for extreme events based on the regionalised fields defined previously. Using the methodology developed in Part 2, it will also be possible to measure the uncertainty of the results by analysing biases and trends in the various models.

Step 2: Study the links between rainfall, river flow rate and floodplain surface area. The floodplains of the Llanos region (Bolivia) have complex dynamics and store large volumes of water. Their surface area is estimated between 70,000 and 150,000 km². However, the water flows seem partially decorrelated from the hydrology of the main watercourses. This calls for better maps of the floodplains and a better understanding of how they relate to the hydrology of the catchment basin. The maps will use MODIS satellite data (daily images of watercourses and floodplains) and altimetric data over the course of several hydrological cycles. Areas at risk for flooding (with different probabilities) will be mapped, for the current period as well as the various climate evolution scenarios, based on the regionalised rainfall fields defined in Part 2.

2) Impacts of climate change in West Africa

The objective is to quantify productivity changes in rainfed tropical crops, which play a major role in the food and financial resources of many developing countries in tropical latitudes. The research will focus on West Africa, where several types of meteorological data (mainly rainfall) and agricultural data (yield, crop area, production, etc.) are available.

Step 1: Identify the most important climate factors for crop yield, via statistical analyses and more mechanistic models that simulate the plant's water needs: SARRAH and ORCH-Mil. The latter model, still under development, is the fruit of a joint effort by the LSCE and LOCEAN to integrate tropical crop representation in the ORCHIDEE model. Future crop yields can be deduced by applying these climate-yield relationships (based on observations) to the regionalised climate forecasts. Uncertainties associated with the climate change forecasts can be assessed using the methodology defined in Part 2. Researchers will be able to study how these uncertainties affect the yield forecasts, in a region where models diverge considerably as to the evolution of monsoon rains, which are decisive for farming. 

Lien entre pluie et rendement des céréales au Niger (bleu : années humides, jaune : années sèches, courbe rouge : rendement céréales)

Relationship between rainfall and grain yield in Niger
(blue: wet years, yellow: dry years, red curve: grain yield) 

Step 2: Develop a model to simulate farmers' behaviour as a function of climate. In this region characterised by highly variable precipitation from year to year, producers prefer to diversify rather than specialise in the most profitable crop, and limit the use of inputs for fear of being unable to reimburse their loans if the harvest is bad. Cotton, corn, sorghum and millet will be the crops studied. The model will simulate the crops and input levels chosen by the farmers and quantify any modifications due to climate change. For the climate change scenarios in which precipitation has been regionalised in a previous step, shifts in crop zones will be estimated by shifts in climate "analogues". It will be particularly interesting to observe whether certain key crops move beyond national borders, potentially causing geostrategic problems.