Data scarcity globally has impeded our understanding of hydrological processes. This study was aimed at evaluating skills of models in reproducing past climate in the Shire River Basin (SRB) in Malawi for future climate impact assessments. The study used data, simulated by Global Climate Models (GCMs), participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5). A total of 52 models were considered comprising a mixture of models in the Representative Concentration Pathways of RCP4.5 and RCP6.0. The mean annual bias, correlation, extreme precipitation indices obtained from the RClimdex package of R software program and frequency distributions were used to quantify the accuracy of the GCM simulations. On the precipitation indices, emphasis was placed on the frequency indices (number of heavy precipitation days (RR ≥ 10 mm), R10mm, number of very heavy precipitation days (RR ≥ 20 mm), R20mm, number of extremely heavy precipitation days (RR ≥ 25 mm), R25mm, Consecutive Dry Days (RR < 1 mm), CDD and Consecutive Wet Days (RR ≥ 1 mm), CWD and on the intensity indices (daily maximum precipitation, RX1day, 5-day maximum precipitation, RX5days, annual total wet-day precipitation, PRCPTOT and very wet days, (R95P). Study results have revealed that there is variation in the performances of individual models and that the overall performance of the models over the SRB is generally low. Some individual models perform better than the multi-model ensemble. Results have also shown the better performance of the following models: ACCESS1-3_rcp45_r1i1p1, BNU-ESM_rcp45_r1i1p1, CSIRO-Mk3-6-0_rcp45_r3i1p1, CSIRO-Mk3-6-0_rcp45_r8i1p1 and GFDL-ESM2G_rcp45_r1i1p1 of medium-low emission pathway, RCP4.5, in replicating the historical extreme precipitation for Shire River Basin.
Research Article
Evaluation of 21st Century CMIP5 GCM Outputs for Climate Change Impact Assessments in Shire River Basin in Malawi
https://doi.org/10.21203/rs.3.rs-418688/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
CMIP5
Global Climate Models
Precipitation indices
Representative Concentration Pathways and Shire River Basin.
Data scarcity globally has impeded our understanding of hydrological processes. The situation is even worse in developing countries such as Malawi. In addition, there is a general agreement by scientists across the globe, that climate is changing and will continue to do so in the future. The extent and direction of change is very uncertain. This development is very worrisome to hydrologists and water managers, in their quest to design and construct hydraulic structures such as barrages, canals, bridges, dams, embankments, reservoirs and spillways. Consequently, there has been a plethora of climate change studies in an attempt to understand the change and come up with proper mitigation and adaptation measures. These studies are periodically done by the Intergovernmental Panel on Climate Change (IPCC), which conducts simulations for future climate caused by different emission scenarios. These studies have culminated into reports such as the Third, Fourth and Fifth Assessment Reports, termed AR3, AR4 and the recent AR5, respectively. In order to properly project future climate change, Global Climate Models (GCMs) are commonly used (Libanda & Nkolola, 2019). Knowledge of how climate will change in the future and the impacts this change might bring, is very vital for planning purposes.
In the recent past, the 21st century climate projections from GCMs participating in the World Climate Research Programme’s CMIP3 have been used to assess climate change impacts at regional and local scales. Currently, CMIP5 is into force and as such, and most simulations are being undertaken using these new generation GCMS. This is because these simulations provide the basis for many of the conclusions in the recent Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5) (IPCC, 2013).
GCM simulations often give a myriad of model outputs, with varying capabilities; intended to accurately reproduce the past and predict the future. Owing to this huge number of outputs, selecting the best performing model has oftentimes posed a big challenge. To avert this challenge, evaluation of model outputs has often been used. Since the future climate change is very uncertain, it is often preferred to test the models in their capabilities to reproduce the past climate and this gives confidence in the models in predicting the future climate.
To understand the performance of CMIP5 GCM simulations, there have been so may evaluations that have been undertaken by different studies with differing objectives. Using 22 CMIP5 historical simulations of precipitation over East Africa (EA), Ongoma et al., (2018) noted that there is huge variation in the performance of individual models and that the overall performance of the models in that region is generally low. These findings are consistent with another study by Rupp et al. (2013), who also noted that there exist differences in the performance of models between regions of the United States. In some studies, it was discovered that model performance was dependent on model resolution (Akurut et al., 2014). This, therefore, requires some modification of the resolutions, if the model performance is to be fairly improved. In some studies, it has been observed that CMIP5 models perform better in some precipitation events including relatively intense events such as wet days (Koutroulis et al., 2016). It must be noted that when it comes to representing observed precipitation on short spatial and temporal scales, GCMs have limits. For instance, Mehran et al., (2014) reported poor reproduction of the observed precipitation over arid regions and certain subcontinental regions across the globe. When evaluating GCM model outputs, which are intended to be selected and applied to climate change impacts, there are always uncertainties. Accordingly, there is need to minimize these uncertainties as much as possible in order to have credible simulations. One way of achieving this is the use of ensemble mean of models. In some cases, the ensemble mean of models have demonstrated close approximation of the mean of the ensemble values with the observation (Nyeko et al., 2011). Woldemeskel et al., (2012) found that the model uncertainties were consistent using a number of groups of models, where ensemble run uncertainty was found to be more important in precipitation simulation than that in temperature. In addition, substantial improvement of the simulation of precipitation was achieved using multimodel ensembles of the CMIP5 and this provided a more realistic fine-scale features tied to local topography and land use (Nikiema et al., 2016). Despite the promise offered by the ensemble means for their reliability in offsetting the downside of individual model runs, there are also some discrepancies in their performance. In some cases, using fully-coupled models of the CMIP5 historical experiments, a study by Yang et al., (2014) observed the underestimation and overestimation of long and short rains, respectively, in East Africa.
Vincent et al., (2014) determined the nature of recent observed climate changes and projected future changes for Malawi, using a combination of GCM ensembles driven by RCP4.5 and RCP8.5 scenarios of IPCC’s AR5. In addition and using 11 GCMs and 21 RCMs, Erika & Reay, (2018) found that the current models are suitable for projecting temperature trends but not precipitation and that future plans will need to consider a range of future precipitation scenarios. Furthermore, using 18 CMIP5 historical simulations of precipitation over Malawi, to examine the performance against rain gauge data, Libanda & Nkolola, (2019) found that in general, most models poorly simulated the spatial standard deviation. In addtion, it was also found that while the selected models were well performing, they were also riddled with a myriad of deficiences. Owing to these findings, modelling improvement was therefore recommended for Malawi. .
Even though the above studies have demonstrated that the evaluation GCMs ability in replicating the observed climate in different regions globally has been conducted, very few have focused on Malawi in general and SRB in particular. The current study focuses on the SRB in Malawi. The SRB lies in the Sub-saharan Africa (SSA) and the SSA has been identified as being particularly vulnerable to future climate change due to its high exposure and low adaptive capacity (Niang et al., 2014). In addition, the SRB’s hydrological system represents Malawi’s most important water resource that provides essential water to key developmental sectors such as hydro-power generation, agriculture, public water supply, fisheries and navigation (International Water Association, 2011).
From the foregoing, the climate change impacts for Malawi and the SRB have been sparesly studied and mostly using the outdated greenhouse gas emission scenarios. In addition, most studies on climate change impacts for the basin, have focused on the averages; no much coverage has been done on the extremes, which are worrisome. Analysis of extreme rainfall events that were analysed at 43 stations across Malawi by Ngongondo et al., (2014) revealed a decrease in total annual rainfall, annual maximum 1-day and 5-day rainfall amount, number of heavy and extreme rainfall days. However, there was an increase in the consecutive number of wet and dry days. Studies on the current latest generation CMIP5 GCM simulations, the ones that motivate this paper, have not been extensively covered, thereby leaving a knowledge gap. It is believed that these latest GCM simulations provide fewer uncertainties while offering better and more reliable predictions. Owing to the identified gaps, this study therefore aims at assessing the ability of the current state-of-the-art 21st century GCM simulations, archived by CMIP5 in reproducing the past climate (precipitation) for the Shire River Basin (SRB) in Malawi. The assessment culminates into the selection of models that fairly reproduce that past precipitation and can therefore be used for future impact assessments.
2.1 Study Area
This study was conducted in the SRB in Malawi (Fig. 1). Shire River is one of the tributaries of the Zambezi River and is the longest and largest river in Malawi. It is home to over 3 million people; flows from Lake Malawi for a distance of about 700km to its confluence with the Zambezi River in Mozambique and 95% of the Shire River is in Malawi and the remaining 5% is in Mozambique. Shire River provides water for more than 95% of Malawi’s hydroelectric power generation, agriculture (e.g., small- and large-scale irrigation schemes in the lower Shire), fisheries, navigation, tourism and urban water supply (90% of the water supplied to Blantyre, Malawi’s commercial capital. The basin has been identified as being particularly vulnerable to future climate change due to its high exposure and low adaptive capacity.
2.2 Data and Sources
This study used two sets of precipitation data namely: the historical (observed) data and CMIP5 GCM dataset.
The historical (control) data was the observed daily data for PCPT of thirty (30) historical years from 1961 to 1990 over the SRB. The precipitation data was averaged from 20 precipitation stations spatially distributed within the study area (Fig. 1). This data was obtained from the Department of Climate Change and Meteorological Services in the Ministry of Natural Resources and Environmental Affairs (MONREA) in Malawi. The CMIP5 GCM-simulated data that was used in this study was also for the thirty (30) years span of 1961–1990 (for control period). This was termed the GCM control simulated data. Table 1 gives detailed information about 30 CMIP5 GCMS which were disintegrated into 52 GCM runs (due to model ensembles) used in this study. A combination of medium-low (RCP4.5) and medium-high (RCP6.0) emission scenarios were used after abandoning GCM outputs from the highest (8.5) and lowest (2.6) representative pathways, owing to their perceived mismatch with the study region.
2.3 Overview of GCM control
This study concentrated on precipitation since it is one of the most important inputs into the hydrological cycle. The assessment of 52 CMIP5 GCM runs was carried out over SRB by comparing the model outputs with observed datasets for 1961–1990 precipitation.
2.4 GCM evaluation approach
The most straightforward method to assess the accuracy of climate models is to compare observed climate data with simulated climate outputs (IPCC, 2013). However, Gleckler et al., (2008) noted that many variables and timescales, which can be taken into account, implies that there is no standard set of tests, which can be applied to a climate model to carry out this comparison. In this study, the annual mean bias, correlations, precipitation indices and frequency distribution were used to evaluate the GCM outputs in their capacity in reproducing the observed climate over the SRB. A total of 52 model outputs driven by medium-low (RCP4.5) and medium-high (RCP6.0) emission pathways, were considered. The extreme pathways of RCP2.6 and RCP8.5 were neglected, owing to their perceived mismatch with the situation in the study area. The GCM control outputs were compared with the observed precipitation for the same period of between 1961 and 1990 (the control period).
In order to measure the inconsistency between the model outputs and the observed dataset, the Root Mean Square Error (RMSE) was used. RMSE has been widely used in climatological and hydrologic studies (Sudheer et al., 2002). Mathematically, RMSE is given as:
where Xobs refers to observed values and Xmodel denotes modelled values at time/place i (Chai and Draxler, 2014).
Where two variables are being compared, there is need to ascertain the degree of change of one variable against another. In order to do this, the strength and direction between the simulated and observed values was established through the determination of correlation where the correlation coefficient, R, was determined (Zaid, 2015).
Considering that this study focuses on extremes, analysis of precipitation extremes, which is the major variable, was done using precipitation extreme indices. The extreme precipitation indices that were analyzed annually were those recommended by the World Meteorological Organization WMO), the Expert Team on Climate Change Detection and Indices (ETCCDI), which has well-defined standard indicators of climate extremes in order to obtain comparable results in different countries. Derivation of these indices was done using the RClimdex package of R software (Zhang & Yang, 2004). Slopes for the simulations were quantified by linear regression (Zaid, 2015). This study used nine (9) extreme precipitation indices (Table 2), categorized into frequency and intensity indices.
When evaluating precipitation extremes, it is sometimes important to determine the optimum fitness of the extreme events by comparing the distribution of the estimates with their corresponding observations of similar return periods. To achieve this, frequency distribution plots were done by using the empirical formula, known as plotting position. From the plots, the quantile versus return period of the simulations were then compared with those of the observations on the same charts. Nyeko et al., (2011) used the same approach in the evaluation of frequency distribution for extremes in the Lake Victoria catchments. To determine the frequency distribution, this study used the Weibull plotting position. For n being the number of years of record and m the rank assigned to the data after arranging them in descending order of magnitude, where the maximum value is assigned m = 1, the second largest value (m = 2), and the lowest value m = n; then the return period Twas computed using the Weibull's formula;
The probability of exceedance of precipitation values was taken as the reciprocal of the return period, same approach used by Kumar & Bhardwaj, (2015).
Since GCM runs have inevitable uncertainties, minimization of these uncertainties is one way of attaining credible outputs. To achieve this, the multi-model mean ensemble for all model runs was computed..
2.5 Model performance and selection
Selection of good performing models is often the end product of model evaluation from a myriad of GCM runs. As this is not a simple task, mean values are often used. Typically, the model runs should have the skills in replicating the past climate accurately. In this study, selection of the best models was done by analyzing the annual absolute biases, correlations, precipitation indices and frequency distribution of precipitation extremes.
3.1 Mean bias
The annual absolute bias for 52 GCM runs for precipitation is shown in Fig. 2. All GCM runs are positively biased, ranging from 9 to 60mm. There is also overestimation of the observed series with the exception of 2 models, CSIRO-Mk3-6-0_rcp45_r3i1p1 (mod25) and MRI-CGCM3_rcp45_r1i1p1 (mod75). These two best performing models are both for the medium-low emission pathways of RCP4.5. Even though the worst performing models such as CanESM2_rcp45_r2i1p1, CanESM2_rcp45_r3i1p1, CanESM2_rcp45_r4i1p1, CCSM4_rcp45_r1i1p1 and CCSM4_rcp45_r2i1p1 are also from the medium-low emission pathways, they are from a different modelling group presumably of different assumption and model structures.
3.2 Correlations
The correlations of GCMs between the observed and the simulated precipitation are given in Table 3. Based on the accepted guidelines for interpreting the correlation coefficient, (Ratner, 2009), the best models that indicate a strong linear relationship between the observed and the simulated precipitation belong to the medium-low emission pathway and these are CESM1-CAM5_rcp45_r1i1p1 (mod17), IPSL-CM5A-LR_rcp45_r1i1p1 (mod56) and ACCESS1-3_rcp45_r1i1p1 (mod1).
3.3 Synopsis of evaluation of precipitation indices
Table 4 shows a summary of slopes of plots for precipitation indices of GCMs considered and Fig. 3 depicts the same. There is evidence of decrease in the dry spell days (CDD), wet spell days (CWD), total annual precipitation (PRCPTOT) and annual maximum 5-day precipitation (RX5). On the other hand, there is observed increase in the number of heavy precipitation days (R10), extremely heavy precipitation days (R25), annual maximum 1-day precipitation (RX1) and very wet days (R95P). Most indices that were analysed were not statistically significant at α = 0.05 level. The above findings are consistent with a study on the analysis of extreme rainfall indices at 43 stations in Malawi where a decrease in PRCPTOT and RX5 was revealed (Ngongondo et al., 2014) and on the analysis of daily precipitation data from 14 south and West African countries over the period 1961–2000, where there was decrease in overall precipitation, average precipitation intensity and increasing dry spell (New et al., 2006).
Both the CDD and CWD of the observations are decreasing by 8 and 6 days per decade respectively (Table 4). The decrease in CWD is less than that of CDD implying that wet days remain wetter while dry days remain drier. The observed slight increase trend in RX1day (0.98mm/decade) is dwarfed by the decrease in RX5day (4.81mm/decade) and the latter seems to have a huge influence on the observed decrease in PRCPTOT. The observed decrease in PRCPTOT of 13mm/decade is coherent with the decrease in CWD and RX5day. Even though heavy precipitation days of R10, R20 and R25 are increasing, this has no major influence on the PRCPTOT, which has demonstrated a relatively large decrease. The increasing number of heavy and extreme precipitation days (R10 and R20) and extreme precipitation days (R25) has potential to bring floods in the region. In terms of the models, most precipitation indices have demonstrated a coarsely equivalent proportion of increasing and decreasing behaviour for all the indicators used in the study area (Fig. 3). The behaviour of R10, R20 and R25 might be explained by intermittent precipitation, which often result to flash floods, which quickly disappear. It is clear from the foregoing that the decreasing PRCPTOT is hugely influenced by the decrease in CWD and monthly maximum consecutive 5-day (RX5day) precipitation. The decrease in PRCPTOT might have also led to decrease in the number of consecutive dry days. This decrease has the potential to cause dry spells for the SRB. The last 5% extreme precipitation series i.e., R95P has shown an increase of 25mm/decade. The increasing number of days for R10, R20 and R25mm has presumably led to this increase. Regarding the simulation skills of GCMs, indices for more than 50% of the models are favourably consistent with the direction of change i.e., decreasing or increasing; the difference being in the magnitude.
3.4 Frequency and Intensity indices
Table 5 and Figs. 4 and 5 summarize of frequency and intensity indices.
Most of the GCM runs are overestimating the dry and wet spells of the observed precipitation as represented respectively by CDD and CWD. However, few models have shown potential to replicate the observed series of all frequency indices. Even though GCMs are over-estimating past precipitation intensity, represented by PRCPTOT, RX1, RX5 and R95P, there is still evidence of a few GCMs’ capability in reproducing the past observed intensity (Fig. 5).
3.5 Model ensembles
Figure 6 shows plots of individual GCM runs, observed mean and the ensemble mean of the simulated GCM runs. There is some over-estimation of the GCM ensemble means to the observed mean and this over-estimation can be described by the presence of some outliers as evidenced in 1963, 1974, 1976, 1978, 1985 and 1989. These outliers might have introduced some uncertainties and in order to achieve a closer replica of the ensemble means with the observations, those outliers can be removed.
3.6 Frequency distribution
The annual frequency distributions of the observations, the simulated and ensemble means, for the 30-year period, are shown in Fig. 7. It has been demonstrated from the plots that even though some GCM runs are overestimating the low, medium and heavy precipitation events, there is still some close mimicking of the precipitation quantiles for all the different precipitation events (return periods). This is consistent with a similar study in Lake Victoria catchments (Nyeko et al., 2011). In addition, the model ensemble has shown evidence of over-estimation of the observed in the first 4 years and under-estimation thereafter. However, the extreme events have been fairly captured by the ensembles.
3.7 Summary of acceptable indicators
See Table 6
3.8 Model Performance, Discussion and Selection
The GCM evaluation has revealed that most models are over-estimating the observed precipitation. This concurs favourably with a study on the hydrological extremes and water resources in Lake Victoria catchments (Nyeko et al., 2011). From the evaluation, the performance of any individual model has not been consistent for all applied indicators. Libanda & Nkolola, (2019) made similar observations in a study on the variability of extreme wet events over Malawi. The divergent performance can be attributed to differences in models’ inherent conditions and internal parameterization made during development each of those models. Additional explanation regarding this behaviour is dependent on models’ sub-grid physics in the atmosphere, ocean, and land domains. To some extent, however, it may also be simply a function of the degree of discretization of the modeling domain. This indicates that the best model(s) for one indicator are not necessarily the same best model(s) for another.
Model evaluation has also revealed that, in addition to over-estimating the observed, there is generally low coherence with observed precipitation. Despite this inconsistency in model performance, models ACCESS1-3_rcp45_r1i1p1, BNU-ESM_rcp45_r1i1p1, CSIRO-Mk3-6-0_rcp45_r3i1p1, CSIRO-Mk3-6-0_rcp45_r8i1p1 and GFDL-ESM2G_rcp45_r1i1p1 have fairly reproduced the observed precipitation and therefore have shown great potential to fairly predict the future climate. Construction of an ensemble of these models would therefore result in reduced performance. This has already been evidenced by the model ensemble findings where it has shown over-estimation of the observed in the first 4 years and under-estimation thereafter indicating over- and under-estimation of the observed low and high precipitation extremes respectively (Fig. 7). Therefore, using a single well-performing model is recommended over ensembles and this was also recommended by Fotso-Nguemo et al., (2018) in another similar study over Central Africa.
This study examined the performance of 52 GCM outputs for precipitation, based on the 21st century simulations of the CMIP5 database. It assessed the capability of those models in reproducing the past daily precipitation for the SRB in Malawi. The objective was to select credible model(s) that can be used for future impact assessment for the basin owing to huge number of GCM outputs that were produced. The control outputs from the GCMs were compared with the observed precipitation during the 1961–1990 (the control period) using a number of performance indicators such as the mean annual bias, correlations, precipitation indices and frequency distribution. This study used nine (9) ETCCDI-defined precipitation extreme indices, selected from a suite of 27 indices, defined by RClimDex.
Study results revealed that there is variation in the performances of individual models and that the overall performance of the models over the SRB is generally low. Some individual models perform better than the multi-model ensemble. This is consistent with a study on the evaluation of CMIP5 20th century rainfall simulation over the equatorial East Africa (Ongoma et al., 2018). Results have also shown the better performance of the following models: ACCESS1-3_rcp45_r1i1p1, BNU-ESM_rcp45_r1i1p1, CSIRO-Mk3-6-0_rcp45_r3i1p1, CSIRO-Mk3-6-0_rcp45_r8i1p1 and GFDL-ESM2G_rcp45_r1i1p1 of medium-low emission pathway, RCP4.5, in replicating the historical extreme precipitation for Shire River Basin. This has been observed in most of the 9 precipitation indices that were analysed. This is coherent with a study conducted in Malawi where projected changes using data from 20 Malawi meteorological stations under the CMIP5 showed the range of projected future changes across 10 different-statistically downscaled CMIP5 GCMs for two different RCP pathways (RCP4.5 and RCP8.5) (Vincent et al., 2014).
This study has offered substantial information on well-performing models and the emission scenario pathway to be used for any study in the SRB. The observed discrepancies or biases in model performance can be dealt with by bias correction and/or downscaling of the coarse data from GCM outputs so as to have reliably data for use in future impact assessments.
Conflict of Interest
The authors declare that they have no conflict of interest.
Funding Statement
The authors did not receive support from any organization for the submitted work.
Author's Contribution
All authors contributed to the study. The draft of the manuscript was written by Petros Zuzani and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.
Availability of data and material
Data and material for this study is available and can be given upon request.
Code availability
Not applicable.
Ethics approval
The authors of this manuscript are committed to upholding the integrity of the scientific record of this journal and as such they have duly followed the Committee on Publication Ethics (COPE) guidelines. In addition, the authors confirm that this manuscript has not been submitted to more than one journal for simultaneous consideration and that submitted work is original and has not been published elsewhere.
Consent to participate
The author gives consent to all co-authors in taking part towards throughout the whole of the research process.
Consent for publication
The authors give consent to publication of this work.
- Akurut, M., Willems, P., & Niwagaba, C. B. (2014). Potential Impacts of Climate Change on Precipitation over Lake Victoria, East Africa, in the 21st Century. Water, 6, 2634–2659. https://doi.org/10.3390/w6092634
- Amhar, U. (2017). Characteristics of Extreme Rainfall in South Sulawesi Province and its Relationship with Sea-Level Temperature Anomalies Around South Sulawesi. School of Meteorology, Climatology and Geophysics.
- Chai T, Draxler RR. (2014). Root mean square error (RMSE) or mean absolute error (MAE)?—Arguments against avoiding RMSE in the literature. Geoscientific Model Development, 7. https://doi.org/7:.10.5194/gmd-7-1247-2014
- Erika, A. W., & Reay, D. S. (2018). Temperature and precipitation change in Malawi: Evaluation of CORDEX-Africa climate simulations for climate change impact assessments and adaptation planning. Science of the Total Environment, 654(2019), 378–392. https://doi.org/10.1016/j.scitotenv.2018.11.098
- Fotso-Nguemo, T., Chamani, R., Yepdo, Z., Sonkoué, D., Matsaguim, C., Vondou, D., & Tanessong, R. (2018). Projected trends of extreme rainfall events from CMIP5 models over Central Africa. Atmospheric Science Letters. https://doi.org/10.1002/asl.803
- Gleckler, P. J., Taylor, K. E., & Doutriaux, C. (2008). Performance indicators for climate models. Journal of Geophysical Research, 113, 2156-2202.
- International Water Association. (2011). The Water Resources Investment Strategy (WRIS)—The Shire River System, Malawi.
- IPCC. (2013). Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press.
- Koutroulis, A. G., Grillakis, M. G., Tsanis, I. K., & Papadimitriou, L. (2016). Evaluation of precipitation and temperature simulation performance of the CMIP3 and CMIP5 historical experiments. Climate Dynamics, 47, 1881–1898. https://doi.org/10.1007/s00382-015-2938-x
- Kumar, & Bhardwaj, A. (2015). Probability analysis of return period of daily maximum rainfall in annual data set of Ludhiana, Punjab. Indian Journal of Agricultural Research, 49(2), 160–164. https://doi.org/10.5958/0976-058X.2015.00023.2
- Libanda, B., & Nkolola, N. B. (2019). Skill of CMIP5 models in simulating rainfall over Malawi. Modeling Earth Systems and Environment. https://doi.org/10.1007/s40808-019-00611-0
- Mehran, A., AghaKouchak, A., & Phillips, T. J. (2014). Evaluation of CMIP5 continental precipitation simulations relative to satellite-based gauge-adjusted observations. Journal of Geophysical Research - Atmospheres, 119, 1695–1707. https://doi.org/10.1002/2013JD021152
- New, M., Hewitson, B., Stephenson, D. B., Tsiga, A., Kruger, A., Manhique, A., Gomez, Bernard, Coelho, C. A. S., Masisi, D. N., Kululanga, E., Mbambalala, E., Adesina, Francis, Kanyanga, J., Adosi, J., Fortunata, L., Mdoka, M. L., & Lajoie, R. (2006). Evidence of trends in daily climate extremes over Southern and West Africa. Journal of Geophysical Research - Atmospheres, 111((D14)). https://doi.org/10.1029/2005JD006289
- Ngongondo, C., Tallaksen, L. M., & XU, C.-Y. (2014). Growing season length and rainfall extremes analysis in Malawi. Hydrology in a Changing World: Environmental and Human Dimensions, 363.
- Niang, I., Ruppel, O. C., Abdrabo, M. A., Essel, A., Lennard, C., Padgham, J., & Urquhart, P. (2014). Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.
- Nikiema, P. M., Sylla, M. B., Ogunjobi, K., Kebe, I., Gibba, P., & Giorgi, F. (2016). Multi-model CMIP5 and CORDEX simulations of historical summer temperature and precipitation variabilities over West Africa. International Journal of Climatology, 37(5). https://doi.org/10.1002/joc.4856
- Nyeko, P. O., Willems, P., & Katashaya, G. N. (2011). Climate change impacts on hydrological extremes and water resources in Lake Victoria catchments, upper Nile basin [Katholieke Universiteit Leuven]. https://limo.libis.be/primo-explore/fulldisplay?docid=LIRIAS1728317&context=L&vid=Lirias&search_scope=Lirias&tab=default_tab&lang=en_US&fromSitemap=1
- Ongoma, V., Chen, H., & Gao, C. (2018). Evaluation of CMIP5 twentieth century rainfall simulation over the equatorial East Africa. Theoretical and Applied Climatology, 19. https://doi.org/10.1007/s00704-018-2392-x
- Ratner, B. (2009). The correlation coefficient: Its values range between +1/−1, or do they? Journal of Targeting, Measurement and Analysis for Marketing, 17, 139–142. https://doi.org/0.1057/jt.2009.5
- Rupp, Abatzoglou, J. T., Hegewisch, K. C., & Mote, P. W. (2013). Evaluation of CMIP5 20th century climate simulations for the Pacific Northwest USA. JOURNAL OF GEOPHYSICAL RESEARCH: ATMOSPHERES, 118(10), 884–906. https://doi.org/10.1002/jgrd.50843
- Sudheer, K. P., Gosain, A. K., & Ramasastri, K. (2002). A data-driven algorithm for constructing artifcial neural network rainfall-runof models. Hydrol Process. Hydrological Processes, 16, 1325-1330. https://doi.org/10.1002/ hyp.554
- Vincent, K., Dougil, A. J., Mkwambisi, D. D., Cull, T., Stringer, L. C., & Chanika, D. (2014). Analysis of Existing Weather and Climate Information for Malawi.
- Woldemeskel, F. M., Sharma, A., Sivakumar, B., & Mehrotra, R. (2012). An error estimation method for precipitation and temperature projections for future climates. Journal of Geophysical Research, 117. https://doi.org/10.1029/2012JD018062
- Yang, W., Seager, R., Cane, M. A., & Lyon, B. (2014). The East African Long Rains in Observations and Models. Journal of Climate, 27. https://doi.org/10.1175/JCLI-D-13-00447.1
- Zaid, M. A. (2015). Correlation and Regression Analysis. The Statistical, Economic and Social Research and Training Centre for Islamic Countries. www.sesric.org
- Zhang, X., & Yang, F. (2004). RClimDex (1.0) User Guide. Climate Research Branch Environment Canada. Downsview (Ontario, Canada).
- Zuzani, P. N., Ngongondo, C. S., Mwale, F. D., & Willems, P. (2018). Examining Trends of Hydro-Meteorological Extremes in the Shire River Basin in Malawi. Journal of Physics and Chemistry of the Earth, (In Press).
Table 1: Information about GCMs used
Table 2: The precipitation indices used (Source: Zhang & Yang, 2004)
Table 3: Correlation between the Observed and Simulated Precipitation
Table 4: Summary of Slopes of Plots for Precipitation Indices of GCMs considered
Table 5: Summary of frequency and intensity indices
Table 6: Summary of acceptable models for each indicator
