Twentieth Century Precipitation Trends in the Upper Mzingwane Sub-catchment of the Northern Limpopo Basin, Zimbabwe


 This study evaluates precipitation trends in the upper Mzingwane sub-catchment (UMS) of Zimbabwe for the variables of annual precipitation, extremely wet days, consecutive wet days and consecutive dry days. The UMS is of strategic socio-economic significance in terms of its provision of water security and sustenance to livelihoods. The analysis is undertaken at four stations: Bulawayo Goetz, Filabusi, Mbalabala and Matopos National Park, and for the period 1921–2000. In general trends were found not to have local statistical significance, with the exception of the Matopos station (in the westernmost extent of UMS) which records significant increasing (declining) trends for most dryness (wetness) extreme indices. A general north to south-western declining precipitation gradient during the past ~ 69 years over the UMS was also found. The findings provide a baseline for future extended historical and future precipitation trend studies, and are important in the context of the socio-economic impacts of extreme events in this region.


Introduction
Pronounced hydroclimatic variability is an intrinsic to climate variability in Africa and often has negative implications for water security (Nash et al. 2016). Furthermore, there is evidence of drastic general warming and regional drying (i.e. decreasing annual precipitation) over large parts of the continent since the start of the twentieth century (Moss et al. 2010; Ramírez-Villegas and Thornton 2015; Miller and Croft 2018; Hannaford 2020). Over southern Africa (SA) for example, analysis of regionally averaged occurrence of climate extremes have revealed declining trends in total precipitation and increase in diurnal temperature range (DTR) coupled with rapid increases in maximum temperature extremes (New et al. 2006; Seneviratne et al. 2012). The decreases in total precipitation have over the past second half of the twentieth century been found to be associated with increasing (though insigni cant) trends in extreme precipitation days and in maximum annual 5-day and 1-day precipitation (Frich et al. 2002).
To better understand future climatic trends and extremes in SA, various Global Climate Models (GCMs) and Regional Climate Models (RCMs) have been used by e.g. (Engelbrecht et (Peterson et al., 2001;Zhang et al., 2011) to explore trends in climate extremes over Africa and globally. However, climate trends remain relatively unexplored in Zimbabwe, especially at sub-national and provincial levels. This has been largely attributed to the paucity of reliable data (Gumindoga et al. 2017), absence of long multi-decadal scale datasets, large data gaps and the closure of meteorological recording stations (Peterson et al. 1998) due to lack of nances and technical expertise.
The available studies on historical climatic trends in Zimbabwe reveal that since 1950, the country has experienced variable changes in climate conditions. Daily minimum (maximum) temperatures rose by ~ 2.6°C ( 2°C) over the last century, coupled with a mean annual precipitation decline of about 5% (Brown et al. 2012 Mazvimavi (2010), found no signi cant trends in extreme precipitation events. The study however was limited to studying extreme occurrences in seasonal and annual rainfall totals, and did not explore the potential impact of arti cial change points in the data. Furthermore, this and other like studies used monthly data to ascertain trends, leaving the analysis of trends in climate statistics at a ner temporal resolution unaccounted for. Such analysis can be undertaken through indices based on daily data, which allow for an objective extraction of information concerning extremes which are known to adversely affect human and natural systems It is against this background that the current study uses historical daily precipitation data to compute WMO -CCL/CLIVAR ETCCDI climate change indices for south-western Zimbabwe. This is the rst such study for Zimbabwe to assess historical (> 50 years) extreme precipitation events and trends using such core indices at a sub-catchment scale. Such an analysis is of particular value to decision and policy makers as it gives a comprehensive understanding of trends in precipitation extremes and their likely current and possibly future impacts on water resources infrastructure and livelihoods. This can further help inform current and future planning and identi cation of the types and range of climate resilience and adaptation options and opportunities to minimise climate extremes exposure risks.

Materials And Methods
We adapt the methodology of Kruger (2006), combined with the change detection test approach developed by Wang et al. (2010). We follow a 5-step approach entailing (1) data collection (digitizing of selected daily precipitation records), (2) pre-processing (cleaning and conversion as explained in Sect. 2.2), (3) testing for arti cial change points in the datasets, (4) further quality checking for errors (i.e. check for duplicates, inconsistent/ unrealistic records), and (5) computation of the ETCCDI indices (Table 1)

Study area
The Mzingwane catchment in the northern Limpopo basin, south-western Zimbabwe, is located in the southern Africa "dry slot" that stretches eastward from Namibia across southern Botswana, southern Zimbabwe and north-eastern South Africa (Engelbrecht et al. 2002). The UMS ( Fig. 1) is one of the four subcatchments of the Mzingwane catchment. It covers 2138 km 2 in areal extent and is located in Natural Region IV of Zimbabwe, which receives ~ 450-650 mm of rainfall per annum (Görgens and Boroto 1997).
This translates to a mean annual run-off of about 600 mm, thus contributing most of the Mzingwane River's run-off. This river in return contributes almost 25% of the Limpopo River's ow volume, and is thus of considerable hydrological importance in the Limpopo basin. Mean annual T max and T min are 26°C and 15°C respectively while potential evaporation ranges between 1800 mm to 2000 mm per annum (Love et al. 2005). The region varies in elevation from ~ 864 to 1560 metres. The ecosystem is typically semi-arid savannah with sparsely distributed woodlands species such as Brachystegia spiciformis, Colophospermum mopane, Terminalia, Acacia, Combretum, aloes, and grass species such as Hyparrhenia lipendula and Heteropogon contortus (Sawunyama et al. 2006).
The UMS extends across three districts: Insiza (to the east), Umzingwane (north-western extent) and Gwanda (to the south-western end) (Fig. 2), and hosts a population of over 50 000 people. It has a diverse agro-ecological and socio-economic structured land

Data collection and pre-processing
Given limited access to daily precipitation data from the Meteorological Services Department of Zimbabwe (MSDZ), datasets utilised in this study were secured from the South African Weather Services (SAWS) library archives and digitised to soft copy format. Unfortunately, post 20th century data are only kept by the MSDZ, in both digital and hardcopy (undigitised format) and can only be obtained at prohibitive considerable expense. Criteria utilised in distilling the datasets were (1) spatial coverage (i.e., stations should be located in the UMS or at least within a 15 km radius from the UMS boundary), (2) have the longest time period possible and (3) retain a su cient number of rainfall records/have minimum missing records. This process culminated in the selection of four stations: Bulawayo Goetz (BG), Matopos National Park (MNP), Mbalabala (Mb) and Filabusi (Fl) ( Table 2), representing the north, east, central and western extents of the study area respectively. Pre-processing was undertaken following the speci cations by Zhang and Yang (2004), where data were converted and structured using the RClimDex software.
Considering that precipitation data series usually contain arti cial shifts that could be related to changes in observation instruments, station location and the environmental setting etc.

Change point detection and data homogenisation
We used the default software parameterisation (e.g. setting the nominal level of con dence at which to conduct the test [p.lev = 0.95]); the maximum number of years of data immediately before or after a change point to be used for estimating the Probability Distribution Function (PDF) (set to use the whole data set without segmentation); and the lower threshold of precipitation (pthr = 0.0). The homogenisation procedure was divided into two primary steps: (i) detection of inhomogeneities (change points) and (ii) calculation and application of data adjustment parameters for each station. The process follows an iterative run of the 'StepSize' function to help detect statistically insigni cant change points and homogenise the dataset by applying the Box-Cox transformation technique. The Box-Cox transformation also lowers False Alarm Rates (FARs) and improves extreme trend detection power (Osborne 2010). Details of this procedure are described in detail by Wang and Feng (2013). Table 3 summaries results for this stage of the analysis, which presents the number of signi cant change points. Given that no change points were identi ed for BG, data were used as is.  Table 1 and details for internal processes of the RClimDex software refer to Zhang and Yang (2004).

Trends and statistical analysis
In order to detect trends, statistical signi cance and the slopes of the index series, we employed the nonparametric Mann-Kendall (MK) trend test (Mann 1945;Kendall 1975) and the Sen's Slope Estimator (Sen 1968) respectively. These have been widely used to evaluate signi cance of trends in numerous studies e.g.

Absolute indices (RX1DAY and RX5DAY)
Trend analysis for RX1DAY and RX5DAY are presented in Fig. 3 and Fig. 4 respectively, while the trend signi cance results are shown in Table 4. RX1Day trends for all but Matopos National Park station were positive and statistically insigni cant. Overall, marginally increasing trends in all the UMS zones (mean = 0.2mm/annum) were found, expect for the western extent which showed a declining trend (mean = -0.463mm/annum) over the study period. Bulawayo Goetz and Filabusi (Mbalabala) stations had insigni cant negative (positive) trends for RX5DAY. At Matopos National Park station a signi cant negative trend (Q med =-0.672, p = 0.018) was recorded. Three of the stations exhibit declining RX5Day trends, (mean = -0.879mm/annum). The negative (signi cant) trends in the western extent of UMS suggest a decline in extreme events in the form of 5-day maximum precipitation.

Percentile-based indices (R95p and R99p)
No signi cant trends were detected for historical very wet day (R95p) and extremely wet day (R99p) events over the study time period for three of the four stations. The exception is MNP, which registered a signi cant declining trend (Q med = -1.41; p = 0.029) for R95p and an insigni cant declining trend for R99p (see Table 4). These declining trends over MNP are consistent with the earlier discussed signi cant declining trend in total annual precipitation (Q med = -3.073, p = 0.011 over the western extent of the UMS marked by a drastic drop in R99p from 150mm to about 50mm between the 1970s and 1990s. Trends for R95p and R99p are shown in Fig. 5 and Fig. 6.

Threshold and other indices (R10, R20 and SDII)
Results for heavy rainfall days above 10mm (R10) and 20mm (R20) per day (Fig. 7 and Fig. 8) reveal that three of the stations record declining but statistically insigni cant trends (mean R10/R20 = 19/9 days respectively), while MNP records signi cant negative trends both indices i.e. R10 (R20) Q med = -0.093, p = 0.023 (Q med = -0.075, p = 0.003). The magnitude of these trends is notably very low (below 1%) at all stations (Table 4). These declining trends in R20 and R10 were matched with declining trends of PRCPTOT (as earlier indicated in Fig. 2) at three of the stations, the exception being at BG which recorded an increasing, albeit insigni cant trend (Q med = 0.626, p = 0.568). Over the same period, trends in the intensity of daily precipitation (depicted by the SDII) signi cantly increased (decreased) for Fl (MNP), Q med = -0.033, p = 0.033 (Q med = -0.035, p = 0.004), while the remaining two stations recorded marginally positive but insigni cant trends.

Duration indices (CDD and CWD)
The indices CDD (CWD) are a measure of dry (wet) 'spell duration' in days. Results reveal Q med (p-values) ranging between − 0.075 and − 0.543 (0.099 and 0.857) for CDD, indicating insigni cant decreasing trends (Fig. 9)  ). However, while there is generally decreasing total annual precipitation (mean = -0.54mm/annum), this is accompanied by increased precipitation intensity (marked by increasing maximum 1-day precipitation amounts (RX1Day)) over most of the study region. While Love et al. (2006) found a general declining trend in annual total precipitation over the entire Mzingwane catchment from years 1931 to 2003, our ndings provide a ner-scale detail on how this declining PRCPTOT trend manifests at a subregional (or sub-catchment) scale for the UMS over the period 1921-2000, including an analysis of extreme events.
In addition, given the fact that trends in precipitation intensity over longer periods (revealed by RX5DAY trends) demonstrate a general declining trend, the northern and interior extents of the UMS indicate general increasing trends in the very wet precipitation index (R95p), while the westernmost extent records a signi cant declining trend. Coupled with this, the UMS has generally experienced marginally positive (statistically insigni cant) trends overall in both R95p and R99p though a declining trend in the westernmost extent is noted for the same indices. There is hence an increasing drying trend over time from north to south-west extent in the UMS which is slightly different from the general north to south gradient reported by authors such as Masvopo (2012) Declining trends in threshold indices (R10 and R20) indicate that there are fewer days with daily precipitation above 10mm (-19/days/decade) and 20mm (-9days/decade) respectively. This may imply that increased precipitation intensities (RX1Day) are concentrated on extreme precipitation days (e.g. R95p) as also suggested in similar studies such as Berhane et al. (2020) in Ethiopia and Kruger (2006)  Results for CWD (wet spells) show a general negative trend in most part of the UMS, apart from Filabusi which has a marginal increasing (statistically insigni cant) trend. The westernmost region (MNP) registered a decreasing trend (~ 2days/decade), hence complementing and elaborating further on the characteristics of the drying trend depicted by indices such as the PRCPTOT for the same western subregion. Further to this, declining trends in the consecutive number of dry days (dry spells) (CDD) were noted in all but the eastern sub-region ( Table 4). The decline in dry spells does not necessarily imply commensurate favourable/ normal precipitation conditions as might be assumed. This could be attributed to the fact that CDD (and CWD) are very sensitive indices, such that even single day rainfall events are enough to terminate a CDD or CWD period as indicated by Hofstra and New (2009). This, however, does not diminish the utility of the CDD index as it has been widely used as a meaningful measure of unusually dry conditions and a potential proxy drought indicator (Gri ths 2007).
An understanding of these temporal changes in climatic extremes and their trends is important given the impacts that extreme events have on society and the natural environment (

Conclusion
Notwithstanding the limitations of access of up-to-date station precipitation data for our study area, we were able to successfully analyse and demonstrate the utility of climate extreme indices in understanding historical precipitation extremes in the UMS for the period 1921-2000. Most computed index values in this study showed negative, statistically insigni cant but spatially coherent precipitation extreme trends across the UMS. The exception is for the westernmost extent of the area, which showed signi cant negative precipitation extreme trends indicative of drying over time. A general shift towards shorter very wet periods with more intense precipitation is noted in the study area. Overall, a general north to southwestern declining precipitation gradient across the UMS is identi ed contrary to the general north to south gradient identi ed in previous studies. Findings from this study could also serve as a baseline in followup studies aimed at extended historical analysis, modelling future trends in extreme precipitation conditions and the likely impacts in the UMS using alternative datasets such as CCAM and CHIRPS.

Declarations
Funding: Partial nancial support (tuition) was received by Auther Maviza from The National University of Science and Technology to undertake this study at the University of the Witwatersrand.

Con icts of interest/Competing interests:
All authors declare that they do not have any con icts of interest to disclose that are relevant to the contents of this article.
Availability of data and material: Not applicable Code availability: Not applicable Location of the four selected climate stations used in this study (marked with purple ellipses) inside and within 15km distance of the UMS Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.