Using Statistical Downscaling of GCM Simulations to Assess Climate Change Impacts on Drought Conditions in the Northwest of Morocco

The present study focus on a statistical approach for improving climate prediction for the region of the Bouregreg basin in the northwest of Morocco. The aim was to examine the past drought conditions and to find their trend under climate change conditions. Past and future drought conditions were assessed using Standard Precipitation Index (SPI) by considering the emission scenarios A2 and B2 for three periods 2014-2040, 2041-2070 and 2071-2099. A statistical downscaling method was applied to the HadCM3 outputs for the station of Rabat to simulate the projected changes in precipitation and temperature. Results shows that annual precipitations under A2 scenario increased by 4.72 %, 4.75% and 4.50 % for the periods 2014-2040, 2041-2070 and 2071-2099 respectively, for the B2 scenario increased by 4.19%, 5.0% and 4.58% for the 2014-2040, 2041-2070 and 2071-2099 periods respectively. The annual mean maximum temperature under A2 scenario increased by 2.24%, 7.43% and 16.31% for 2014-2040, 2041-2070 and 2071-2099 periods respectively, and for the B2 scenario increased by 3.07%, 6.97% and 10.0% for 2014-2040, 2041-2070 and 2071-2099 periods respectively. The results showed also an overall decrease of annual and seasonal drought severity over the years. The annual and seasonal drought severity and duration for the periods 2025-2045 and 2055-2070 will increase under A2 and decrease under B2 scenario.


Introduction
Drought is a natural hazard that can have serious consequences for a range of human activities.Its impacts are more felt economically and socially: agricultural production and water resource availability for industry and households being the most affected sectors, particularly in areas of the world that do not have the infrastructure to effectively mitigate its affects.Drought is often seen as a "creeping" phenomenon with slow onset and cessation.As a result, an effective drought monitoring system is the most important tool for developing and implementing efficient mitigation strategies.However, not only can the onset of drought conditions be rapid, an indication of how long drought conditions may continue will enable improved planning and resource allocation.For this reason, a capability to accurately forecast the onset, persistence and cessation of drought conditions will enable more effective drought mitigation strategies to be developed.
Drought can be defined as a period of time with water availability less than some specified amount at a particular location.It is primarily driven by a shortage of precipitation, the effects of which can be enhanced or reduced at any stage of the water cycle.Therefore, as a means towards developing a drought forecasting system, this study concentrates on forecasting the temperature and precipitation contribution towards drought conditions.
The rainfall distribution in Morocco is very scarce and characterized by high annual variability.Several economic sectors depend on the average annual rainfall like agriculture and tourism.Therefore, economy become very sensitive to climate changes and thus water management projects are highly requested.The study of rainfall distribution and the future trends is an essential task to better planning and managing water resources.Some studies associate rainfall variability in the northern of Morocco to the North Atlantic Oscillation (Knippertz et al.2003a) which have a major impact on winter climate.High Moroccan precipitation tends to coincide with large negative anomalies of the NAO while drought in Morocco is related to a positive state of NAO.The need of regional understanding of precipitation variability in northern Morocco seems to be very important to better forecast drought occurrence and develop efficient mitigations methods.
General Circulation Models (GCMs) have recently been deployed for weather forecasting and climate change projections, and can reasonably represent the large-scale aspects of climate, but they do not provide enough prediction information for the local and regional climate because of their coarse resolution of several hundreds of kilometers (Juneng et al. 2010).
Statistical downscaling is a two step process consisting of the development of statistical relationship between local climate variables like "precipitation" and large scale predictors like "pressure field" and the application of such relationships to the output of GCMs to simulate local climate characteristics in the future.Statistical downscaling is used in climate impact assessment at regional and local scales and when suitable observed data are available to derive the statistical relationships.

Study Area
The target area of this study is the region of Bouregreg basin in the northwest region of Morocco (figure 1) (32°06'-35°56' N, 3°46'-8°01' W) which covers a total area of 62 960 km².The Bouregreg basin covers an area of 9771 km².It's altitude range from sea level to 1724 m.The climate is semi arid influenced by altitude, latitudinal extension and the opening of the region on the Atlantic costs.Precipitations recorded in area with high altitude (Northeast and Southeast) are higher than in plain.Latitudinal extension leads to a climatic pattern temperate in the North and semi arid in south.Region exposed to Atlantic Ocean are more wetted with moderate temperatures.The mean annual temperature is 18 °C and the total average annual precipitation is 440 mm. the precipitation decrease slightly with the latitude and range from 480 mm/y in Rabat to less than 370 mm/y in the southwest of the basin.However its show a strong increases with altitude and reach 760 mm/y at the highest location.The mean annual evapotranspiration is 1600 mm in the coastal region of Rabat and 800 mm in the high region of Bouregreg.We focus in this study to the station of Rabat which has a considerable historical quality data.

Figure 1. Localisation of the northwest region of Morocco
In this study we have used the HadCM3 model simulations, where the outputs were downscaled by the Automated Statistical Downscalling (ASD) model (Hessami et al. 2008) inspired by the existing Statistical Downscaling Model (SDSM, developed by Wilby et al., 2002).Figure 2 shows the general scheme of the ASD algorithm for generating climate scenario information.Data used for predictors derived from the National Centers for Atmospheric Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) that have accomplished different re-analysis projects which aim on the generation of global data sets for a long time period for different atmospheric parameters.The re-analysis is created with a model similar to the one used for weather forecasts.This model is initialized with measured data from different sources, including observations from weather stations, ship, aircraft, radiosonde, and satellite.Using one model for the whole re-analysis period generates homogeneous data that can be used for long term studies (Kalnay et al. 1996).

General Circulation Model (GCM) Output
For future projection we used data from the HadCM3 model which is a coupled climate model that has been used extensively for climate prediction, detection and attribution, and other climate sensitivity studies.HadCM3 was one of the major models used in the IPCC Third and Fourth Assessments, and also contributes to the Fifth Assessment.Its good simulation of current climate without using flux adjustments was a major advance at the time it was developed and it still ranks highly compared to other models in this respect (Reichler & Kim, 2008).
It also has the capability to capture the time-dependent fingerprint of historical climate change in response to natural and anthropogenic forcings (Stott et al. 2000) which has made it a particularly useful tool in studies concerning the detection and attribution of past climate changes.The scenarios chosen for future climate scenarios are A2 and B2: A2 scenario: Regional heterogeneous development: High population growth.Economic development primarily regionally oriented.Economic growth and technological change are more fragmented and slower than in A1.

B2 scenario:
A world in which the emphasis is on local solutions to economic, social, and environmental sustainability.Moderate population growth, intermediate economic development, less rapid and more diverse technological change.Focussing on local and regional levels.
In order to finalize the downscaling method we retrieved predictor data from HadCM3 model output, for both A1 and B2 scenarios emission, at the nearest grid box (35, 352.5) of Rabat station.

Standard Precipitation Index (SPI)
The Standardized Precipitation Index (SPI) has been defined as a key indicator for monitoring drought by the World Meteorological Organization (WMO, 2012) and has been widely applied as an operational (Wilhite et al., 2000;McRoberts & Nielsen-Gammon, 2011) and analysis tool.The SPI is a probabilistic indicator based purely on precipitation.The SPI was introduced by McKee et al. (1993) as measure of the precipitation deficit that is uniquely related to probability.It can be calculated for any accumulation timescale, usually from monthly precipitation observations, and is typically expressed as SPI-n, where n is the number of months of accumulation.
The time series is analogous to a moving average in the sense that a new value is calculated each month and is auto-correlated to previous months depending on the accumulation timescale.The computation of SPI is based on an equi-probability transformation of the probability of observed precipitation to the standard normal variable with mean 0 and variance 1. SPI is therefore expressed in units of the number of standard deviations from the mean, with negative (positive) values denoting drier (wetter) conditions than "expected" for the timescale and location.The standardization procedure to the standard normal variable means that the SPI is spatially and temporally invariant.This characteristic enables precipitation anomalies to be objectively compared between locations and times.
The SPI is defined as:

=
Where SPI i is the standardized precipitation index in the period i; Xi is the precipitation for the period i; is the mean precipitation in the period i for the historical series; Sd is the standard deviation of the mean precipitation in the period i.The SPI index was calculated using observation data for the period of 1961-2001.For future periods 2014-2040, 2041-2070 and 2071-2099 we used the monthly time series for precipitation and temperature derived from HadCM3 output (2.5° latitude, 3.75° longitude).To retrieve data for the Rabat station we extract the data of the nearest (latitude, longitude) grid box: (35, 352.5).

Temperature and Precipitation Projections
We have used NCEP daily data for the station of Rabat, from 1961 to 1985 and from 1986 to 2001 respectively to calibrate and validate the ASD model.The future monthly precipitation and temperature were simulated using the HadCM3 model with the A2 and B2 scenarios.The RMSE values for the monthly mean minimum and maximum temperature vary between 0.1 -0.78 for calibration and between 0.51 -1.16 for validation.This showed that ASD model reproduce reasonably the monthly minimum and maximum temperature.
Precipitations simulated are close to observations, but its have more accuracy under B2 scenario than A2.The precipitation projected was overestimated in winter and spring months (except January) and underestimated in autumn.The ASD model simulates temperature more accurately than precipitation.The HadCM3 outputs for monthly mean maximum and minimum temperature and precipitation were used for assessing climate changes impacts on drought severity and occurrence for the periods: 2014-2040, 2041-2070 and 2071-2099, under the A2 and B2 scenarios.(Figure 3) The projected mean maximum and minimum temperature under the A2 and B2 scenarios will increase, especially for the 2071-2099 period.Using B2 scenario the difference between observed and projected data was lower than under A2 scenario in the calibration and validation.Increasing maximum and minimum temperature under A2 scenario is higher than under the B2 scenario.The range of the maximum temperature under A2 varies from 2.5 °C to 4.6 °C.For winter is 2.8 °C to 4.6 °C, for spring is 2.5 °C to 4.2 °C, for summer is 3 °C to 3.9 °C and for autumn is 3 °C to 4.5 °C.Under B2 it varies from 1.3 °C to 2.8 °C, and for the four seasons it range from 1.5 to 2.8 °C, 1.3 to 2.3 °C, 1.9 to 2.6 °C and 2.3 to 2.8 °C in winter, spring, summer and autumn respectively.
The range of the mean minimum temperature under A2 varies from 1.7 °C to 3.0 °C and 0.9 to 2.1 °C under B2.

Past Period
The monthly rainfall in Rabat for the period 1961-2010 is represented in the figure 4. It shows the variability of precipitation with an alternation of shorter or longer dry and wet episodes.Figure 5 shows the SPI time series for different month's steps.The minimum value of SPI is observed in December of 1974.The analysis of the 6 and 12 months SPI series reveals a frequent drought during 1966-1967, 1980-1982, 1993-1995 and 2007-2008.The year 1994 of the dry period of 1993-1995 have the smallest annual precipitations of the study period.The number of dry and wet months and duration of drought and wet periods for the study period is shown in Table 3.

Future Trends
We use the projected SPI index under A2 and B2 scenario (Figure 6) to estimate future drought severity for Rabat region.Visual inspection of SPI time series at 1, 3, 6, and 12 months showed an overall decrease of annual and seasonal drought severity over the years.The annual and seasonal drought severity and duration for the periods 2025-2045 and 2055-2070 will increase under A2 and decrease under B2 scenario.The scenario A2 predict higher severe drought than B2 scenario.
Figure 6.Future projections (for A2 and B2 scenarios) of SPI time series for Rabat at 1, 3, 6, 12 months time step

Conclusion
A study of historical and future precipitation and temperature for the region of the Bouregreg basin is presented in this paper; we used the HadCM3 model outputs to estimate the future trends of precipitations and temperature for the periods: 2014-2040, 2041-2070 and 2071-2099.This study showed an increase in mean temperature and precipitation for the region of Bouregreg basin in the northwest of Morocco as a result of increasing greenhouse gases in the atmosphere.The emission scenarios A2 and B2 from IPCC have been used.The results of this study showed an increasing trend for the projected annual mean precipitations and temperature under both emission scenarios A2 and B2 for the three study periods.The annual mean minimum temperature will increase by 2.1 °C and 1.4 °C for the period 2071-2099 under A2 and B2 scenarios respectively, while the annual mean maximum temperature will increase by 3.6 °C and 2.2 °C under A2 and B2 for the same period.The analysis of drought occurrence and severity showed an increasing of annual and seasonal drought severity and duration for the periods 2025-2045 and 2055-2070 under A2 scenario and decreasing under B2 scenario.The results of this study showed that drought duration and severity in Morocco northwest region will be affected by climate changes,

Figure 2 .
Figure 2. Algorithm of the ASD model

Table 1 .
Predictor used in the downscaling process derived from CCCSN

Table 3 .
SPI statistics for the study period