# «DIPLOMA THESIS Linking Climate Change with Food Security in the highlands of Khyber Pakhtunkhwa, Northwest Pakistan Presented by: Martin Kienzler ...»

To get a higher resolution climatology for the study area on the one hand and a database for future prediction for the 21st century on the other hand, additional to the CRU dataset the output of the regional climate model REMO was used. According to Mannig et al. (2012) regional climate models are “useful tools for prediction of climate change on regional scales”. Because they are restricted to certain regional areas they enable high-resolution long-term simulations. But this means also that information about the lateral and lower boundary conditions are required from GCM simulations or observational data sets. In this case the lateral boundary forcing is given through the ECHAM data. Although boundary eﬀects can occur at the three to eight outermost grid cells of the domain. The “artiﬁcial lateral boundaries of the regional model” thus “form the most critical disadvantage of such models” (Jacob & Podzun, 1997). REMO exists in several resolutions and for several regions and generally is a three-dimensional hydrostatic atmospheric circulation model. The model run which is used for this study is REMO (ECHAM, 1/6◦ ), generated for Central Asia. It is a two step nesting product derived from REMO (ECHAM, 1/2◦ ), which on its part is driven by a run of the coupled General Circulation Model (GCM) ECHAM5/MPI-OM (Roeckner et al., 2003).

The horizontal resolution of the simulation is 0.16◦ x 0.16◦ and covers an area extending roughly from 25◦ North to 45◦ North and from 55◦ East to 85◦ East, counting 181

** Chapter 4. Climate data**

x 121 grid boxes. But the real model domain is not rectangular because the grid box centres are deﬁned on a rotated coordinate system. The region used for this study is an extract of the whole area and covers the same area as for the CRU data (see section 4.1), except that the region between 23.5◦ North and around 25.5◦ North is not represented.

However, this is irrelevant for this thesis, because the main focus area is covered entirely by the grid of the dataset (Jacob & Podzun, 1997; Jacob, 2001; Mannig et al., 2012). The temporal span is 1961 to 2100, which makes it possible to produce future climate predictions with the help of the REMO dataset. The forcing for the period 1961 to 2000 is a standard scenario of observed greenhouse gas concentrations. For the years 2000 to 2100 the Special Report of Emission Scenario (SRES) A1B (Nakicenovic & Coauthors, 2000) is used as forcing.

** Figure 4.3 – The topography of Pakistan and the location of the case studies.**

REMO model output Mannig et al. (2012) compared the output of the REMO (ECHAM, 1/6◦ ) simulations with the CRU and reanalysis datasets for model validation. The result is that the CRU data has the smallest temperature range in the Central Asia region. This is referred to the statistical interpolation method and the relatively sparse distribution of stations. In contrast the REMO data shows the coldest mountainous areas and the warmest valleys, which is attributed to the higher resolution of the applied elevation model. Topographic features and even small-scale climatic controls such as luv and lee eﬀects are taken into account. Regarding precipitation patterns the advantages of the dynamical downscaling of REMO become apparent. The heterogeneous pattern of

** Chapter 4. Climate data**

mountain precipitation is representated distinctly by the highest resolution REMO simulation and it seems that it agrees a lot more with the observational datasets than the reanalysis data does. On the other hand REMO simulates less precipitation in some dry areas and too much rainfall in mountainous areas compared to observational data. Furthermore it seems that the ECHAM driven REMO simulations have a wet bias in regions with a summer precipitation maximum throughout the year and somehow overestimate winter precipitation in dry areas. Regarding temperature REMO simulates a delayed seasonal cycle in most regions. This is the result of a larger snow cover than in reality due to overestimated winter precipitation. Hence the energy required for sensible heat ﬂuxes in spring is used for snow melting in the model. Generally the simulated climate patterns “are in good agreement with observations”, while the interpolation methods of these “tend to smooth the climatological patterns” (Mannig et al., 2012). Similarly Jacob & Podzun (1997) validated precipitation amounts of the Indian summer monsoon simulated by a REMO model against a long-term observational climatology. They, too, claim that REMO is able to reproduce realistic precipitation patterns quite well.

These validation results show that the REMO output is adequate to be used for climatological studies on a regional scale. For this reason it is relied on where the CRU data cannot represent the climatology properly. Especially in the high-mountain surroundings of the two case studies, where small-scale topographic diﬀerences are extreme, the climate situation is represented better by the REMO data.

5 Methodological approach

5.1 Trend analysis The main focus of this thesis is to obtain potential changes of climate within the borders of Khyber Pakhtunkhwa. In a ﬁrst step observed changes shall be detected for the period 1901-2009. In a second step projected changes shall be estimated for the next century until 2100.

The climate of a region is characterized by several climate variables. The most important are temperature, precipitation, air pressure and humidity. In this study only the two variables temperature and precipitation are taken into account. These are variables which represent relatively reliable values and are highly available in climate records. For this study mean values of the two variables are used.

In this thesis annual mean values as well as seasonal mean values are used for trend analysis to detect eventual diﬀerences of changes in the diﬀerent seasons, which is especially important for agriculture. For example in monsoon inﬂuenced regions summer precipitation is the most demanding factor for cropping. In contrast in semi-arid areas where summer precipitation is almost zero, rainfall during winter is more important. For this reason the focus of this analysis is laid on seasonal values. The seasons are deﬁned

**after the standard meteorological deﬁnition:**

• Winter: December of the previous year, January, February (DJF)

• Spring: March, April, May (MAM)

• Summer: June, July, August (JJA)

• Autumn: September, October, November (SON) In climatology calculation and trend analysis is usually referred to a period of 30 years.

These climate normals (CLINO) of diﬀerent climate elements, which are consulted for characterization of the climate for a certain region or place are mostly based on the periods 1901-1930, 1931-1960, 1961-1990 etc. (Schönwiese, 2006). For this study trends are being calculated for the period 1971 to 2000. This enables a comparison of the CRU and REMO datasets and describes the most recent period. The trends are extrapolated

to a 100 year period. For the CRU data a 109 year trend is being calculated additionally for the period 1901 to 2009. Furthermore trends are being built for the periods 1901 to 1940 and 1941 to 1970 to compare the individual sections of the twentieth century.

whereat xi, i = 1,..., n (n = amount of values) are the data, in this case temperature and precipitation values. On the basis of this equation mean values for the two variables and the diﬀerent seasons are calculated.

Variance For trend analysis two more statistical indices have to be calculated: variance and covariance. The variance is one of the moments of a statistical distribution. It basically describes how far a set of values is spread out. It is calculated from the standard deviation, which is a measure of the variability or diversity of a sample. The standard deviation

Covariance The covariance basically is a measure of the strength of the correlation of two random variates. The covariance of two variables xi and xj is deﬁned as

(Schönwiese, 2006) whereas xi, i = 1,..., n and xj, j = 1,..., n represent the two variables. In this case xi is the time steps and xj is the data of the climate variables, namely temperature and precipitation. The range of values (n) for this case is 30.

For uncorrelated variables the covariance is zero. But if two variates correlate in some way, then their covariance is nonzero. If both variates xi and xj are increasing, their covariance is positive: cxi,xj 0. If one variate xi is increasing, while the other variate xj is decreasing, their covariance will be negative: cxi,xj 0.

5.1.3 Trend analysis Linear regression of a time series According to Schönwiese (2006) a time series is a collective of data or a sample {xi } of the extent n (i=1,...,n). Its numerical values apply to discrete points of time ti. In general the data of a time series should be equidistant, which means that a constant time

**step ∆t = ti+1 − ti is used. So a time series has the following mathematical form:**

xi (ti ); ti+1 − ti = ∆t = const.; i = 1,..., n (5.9) A trend analysis of a time series technically is a correlation of the data xi = xi (ti ) with time t, more precisely with the assigned points of time ti or the time steps ∆ti respectively. It is assumed that no other factors play a role for this relation, so the model of linear regression is valid. In this case only one explanatory variable (t) denotes the scalar variable (xi ).

That means that the equation of the regression line of the variables x (dependent) and t (independent)

Trend Thereby the trend of a time series is nothing else than the slope of the regression line B.

This is deﬁned as the covariance of the two variates divided through the variance of the

**cause variable (the time t in this case):**

at which n again is the extend of the time series and amounts to 30 years. It refers to the period 1971-2000 for the REMO data. For the CRU data the trend was calculated for the periods 1901-1940 (n=40 ), 1941-1970 (n=30 ), 1971-2000 (n=30 ), 1981-2009 (n=29 ) and 1901-2009 (n=109 ).

The characteristics of symmetry is not valid anymore for equation 5.14.

Residuals The residual i is a measure for the deviation of a sample from its theoretical value and is important for the procedure of the hypothesis testing (see section 5.1.4). It is calculated as

5.1.4 Student’s t-test A statistical t-test generally is a hypothesis test in which the test statistics follow a Student’s t distribution. This is valid if the null hypothesis is supported. In this study a t-test is applied to ﬁnd out if the slope of the regression line B is signiﬁcant or not.

The null hypothesis H0 describes that there is no slope at all, which means that there is no inﬂuence of the regressor t on the variable x. In contrast the alternative Hypothesis

**shows that the slope is either positive or negative:**

5.1.5 Diﬀerence of means To evaluate possible changes for the 21st century with the help of the predictions of the REMO simulations, the easiest method is to compare mean values of the two climate variables temperature and precipitation for the end of the 21st century and a recent period. For this purpose the periods 2071 to 2100 and 1971 to 2000 were selected. Both periods represent a time span of 30 years and are 100 years apart. The diﬀerence of the means is calculated by subtracting the mean of the period 1971-2000 from the mean of

**the period 2071-2100:**

5.2 Qualitative research: semi-structured interviews For supporting the statement of the analysis of the climate data qualitative interviews were conducted with farmers inside the study area. Qualitative studies are kind of a “nonnumeric approach to research” and “are typically small-scale intensive pieces of research” (Philip, 1998). According to him, the required “information or data may be obtained from a variety of methods”, including participant observation, group interviews, unstructured interviews, structured interviews or questionnaires, to mention only a few.

Bernard (2011) describes another method, which is a mixed form of structured and unstructured interviews: semi-structured interviews. This method requires an interview guide, which is a “written list of questions and topics that need to be covered in a particular order” (Bernard, 2011). The advantage of this is the high ﬂexibility that could give the interviews new leads and ideas but still assures the possibility to get “reliable, comparable qualitative data”. For this reason this method was found to be the adequate one for this study. The reason why qualitative research was preferred to quantitative methodology simply is availability of time, accessibility of the areas of the case studies and therefore the insuﬃcient amount of informants for a quantitative analysis. Furthermore the theory developed through a qualitative approach often is “developed after the information or data have been gathered” (Philip, 1998).

The questioning targets the perception of the local population with respect to the

**subject matter. Principally the interviews focus on two main topics:**

• Do the farmers already realize changes in climate over the last few decades?

• Are there occurring changes in agriculture which can be related to climate change?