Seasonal, Decadal, and El Niño-Southern Oscillation-Related Trends and Anomalies in Rainfall and Dry Spells during the Agricultural Season in Central Malawi

Abstract

As governments continue to address climate change when formulating policy, there remains a need to determine if such a change exists in the historical record to inform clear indices for monitoring the present climate for site-specific interventions. This study characterised trends and anomalies in rainfall and dry spells, providing local information often projected from satellites or regional data in data-scarce regions. From 1961 to 2007, daily rainfall records in Central Malawi were used to calculate indices for low-(Balaka), medium-(Bunda, Chitedze, KIA), and high-altitude (Dedza) sites, which were then subjected to Mann–Kendall’s, Cramer’s, and Spearman-Rho’s trend tests. Significant decreasing trends in terms of wet days and growing season length were evident across locations. Seasonal and extreme rainfall, dry spells, and inter-seasonal and near-decadal anomalies were not consistently or inevitably significant. Unexpectedly, rainfall anomalies were largest in Bunda and KIA, which have mild climatic regimes, while the lowest were in Balaka, a rainfall-averse zone. The relationship between El Niño-Southern Oscillation (ENSO) and extreme rainfall and dry spell events did not reach statistical significance. In conclusion, extreme precipitation and dry spell events show varied intensities and proportions rather than increased frequency. The disparate results largely justify the need for in-depth local-scale assessments for agroclimatic applications.

1. Introduction

Agricultural productivity in Southern Africa is significantly challenged by marginal and unpredictable rainfalls, alongside high runoff and evaporation losses [1]. Climate-smart agriculture and similar sustainable land management technologies have been promoted to enhance soil water availability, which is especially beneficial during periods of low or unpredictable rainfall [2]. However, the adoption of these technologies has not been rigorously assessed across different agroecological zones, which vary widely in terms of rainfall patterns and characteristics. Therefore, understanding local rainfall variability is essential for developing effective, location-specific adaptive agricultural resource management strategies.

Malawi’s diverse climate and geography, influenced by its length, topography, and proximity to Lake Malawi, vary from lowlands to plateaus over 1500 m and mountains up to 2500 m [3]. This diversity complicates the identification of climate trends amidst environmental noise. Research on precipitation trends, which is crucial for understanding climatic variability and change, involves developing indices for planting dates, rainfall patterns, and intensity, essential for rainfed agriculture [4]. Over the past 4 to 5 decades, work on rainfall trends in Malawi has produced varied findings, with no clear evidence of long-term changes in rainfall totals, onset, or cessation times [5,6,7,8,9]. These inconsistencies are attributed to Malawi’s diverse geography and elevation, situated within the Great East African Rift Valley, where both tropical and extratropical processes influence precipitation variability [8,10]. This scientific uncertainty contrasts with local perceptions of decreasing rainfall and shifting seasonal patterns, complicating adaptation efforts [11]. The mismatch between empirical data and local experiences further challenges the development of effective, targeted adaptation strategies, leading to inadequate responses and a focus on short-term coping mechanisms over sustainable, proactive planning [7,12].

Research has shown both year-to-year and multi-decadal rainfall variability, often analysed through normalised rainfall departures to identify periods of abnormal rainfall [13]. Alexander and Nyasulu [14], using data from nine reference stations, observed a shift from wet years between 1978 and 1990 to predominantly dry years from 1991 to 2010 in Malawi. However, our study further contributes to the discourse on interannual and decadal variability by exploring how, at local scale, large-scale climate phenomena, such as the El Niño Southern Oscillation (ENSO), influence such patterns. Rainfall is often below average during Pacific warm ENSO episodes and above average during Pacific cold episodes or the La Nina. Studies suggest that the ENSO may account for up to 50% of the interannual rainfall variance in the region [10,15,16]. Cook et al. [17] demonstrated through the use of simulations that a causal relationship between ENSO events and maize yields exists in Kenya, Malawi, South Africa, and Zimbabwe. However, they observed that ENSO teleconnections to the hydroclimate are generally weaker at finer scales compared to broader geographic regions.

The most precise techniques for analysing climate trends are considered to encompass General Circulation Models, statistical downscaling, averaging station data across broad areas, and employing climatological or geographical divisions [8,18,19,20]. However, at regional or national levels, trends in extreme precipitation often appear spatially incoherent. The limited data from sparse gauge stations and the complex interactions between land and climate can lead to unrepresentative climate analyses [21]. Haylock and Nicholls [20] noted that without normalising the averaging of multiple data series, those with higher means or variances could skew the results and affect the statistical significance of the observed trends.

Additionally, studies indicate that rainfall variability is more pronounced at smaller scales, such as within catchments than over broader spatial and temporal dimensions. Kiely [22] observed up to a 30% reduction in rainfall, while increases in precipitation were also evident. This observation was based on five decades of precipitation data from eight sites and streamflow data from four sites in Ireland. Burn and Elnur [23] found significant variability in sixteen hydrological variables across Canadian catchments. Recent work in Malawi emphasises the need for ground gauge measurements to corroborate findings of notable precipitation changes derived from high-resolution datasets and various rainfall variability indexes [24]. Moore et al. [25] argued for more detailed data at finer scales and the capability to identify processes that explain any discrepancies found at broader measurement scales.

Our study analysed precipitation and dry spell variability, seasonal and decadal trends, and ENSO-driven anomalies in Malawi from 1961 to 2007 using daily data from stations located in three key agroecological zones. It evaluated the frequency, length, and intensity of various seasonal and extreme weather indices, including percentile, absolute, threshold, and duration measures, to determine their relationship with long-term climatic shifts and to identify significant trends. Notably, it introduced three percentile indices—extreme intensity, frequency, and percent—previously unstudied in Malawi. The purpose is to enhance the availability of data that support the development of resilient crop and land management practises amidst changing rainfall patterns. The paper is organised as follows: section two describes an overview of the study area’s physical and climatic context and data sources; section three refers to the methods; section four details the results obtained; and the discussion and conclusions are presented in sections five and six, respectively.

2. Study Area and Data

2.1. Study Locations

Malawi is located in the Great East African Rift Valley, spanning from 9°22′ to 17°7′ S latitude and from 32°40′ to 35°55′ E longitude (Figure 1). The elevation varies across different regions, from 37 m above sea level (MASL) at the southernmost point of Malawi to over 1000 MASL in the mid-altitude plateaus, reaching up to approximately 2500 MASL in the mountainous areas, with Mt. Mulanje in southern Malawi peaking at 3000 MASL [3]. Therefore, the country exhibits significant geographical variability and a wide range of climates. The climate regime is also influenced by Lake Malawi, which spans two-thirds of the country’s length to the east, as well as the Inter-Tropical Convergence Zone (ITCZ), which shifts north and south across the African continent with the seasons [5,26]. The warm and wet season lasts from November to April as the ITCZ moves southward, bringing rainfall.

These physiographic features influenced the selection of three major climatological and geographical zones in Malawi, each represented by the meteorological stations listed in

Table 1. List of meteorological stations and their geographical features.

No.	Station Name/Type	WMO ID	Latitude (S)	Longitude (E)	Altitude (MASL)	Mean Annual Temperature (°C)	Mean Annual Rainfall (mm)
1	Balaka, gauge	-	14°98′	34°97′	625	25	801
2	Dedza, reference	67,689	14°32′	34°25′	1590	16	970
3	Bunda, gauge	-	14°18′	33°77′	1118	21	1030
4	Chitedze, reference	67,585	13°97′	33°63′	1149	20	868
5	KIA, synoptic	67,586	13°78′	33°78′	1229	21	944

. The mid-altitude zone was represented by the Lilongwe north, central, and south stations, specifically Kamuzu International Airport (KIA), Chitedze Agricultural Research Station (Chitedze), and Bunda College of Agriculture (Bunda), respectively. The high-altitude zone was represented by the reference station at Dedza, while the low-altitude zone was represented by the gauge station at Balaka. The mid-altitude stations are located within the Lilongwe Plain, covering 6.5% of Malawi’s land and ranging between 1000 and 1400 MASL [27]. Dedza, at the base of Central Malawi’s highest point—the 2200-MASL Dedza Highlands—experiences significant climatic influences from both the highlands and Lake Malawi. Balaka, in southern Malawi, spans 2.5% of the national territory on lowland plains and isolated hills, with elevations from 350 to 800 MASL. The locations of these stations are depicted in Figure 1.

2.2. Rainfall Data

Daily rainfall datasets, pre-screened and covering the period from November to May across 47 years (1960/61–2006/07), were obtained for recording stations at KIA, Chitedze, Bunda, Dedza, and Balaka under the auspices of the Malawi Department of Climate Change and Meteorological Services (DCCMS) and the Famine and Early Warning Systems Malawi programme. These datasets underwent several basic quality control procedures, such as data smoothening and normality or independence distribution criteria, performed by the agencies in accordance with World Meteorological Organisation standards. Several studies have performed climate assessments using the same datasets and overlapping periods, validating their quality for various analyses [18,28,29]. Datasets that span 30 years and above are considered optimal for climatological studies, whereas the period from 1961 to 1990 is designated as the reference epoch for climate change-related studies [30,31]. This period also falls within the 1948–1988 interval, which is consistent with the most rapid apparent global warming phase of the 19th century [32].

3. Methods

3.1. Rainfall Indices

The indices used in the study and listed in

Table 2. Description of selected seasonal (November–May) and extreme rainfall indices (DP = daily precipitation amount).

Index Name	Description	Units
Threshold indices
Wet days (WD)	Seasonal total number of wet days (DP > 0.85 mm)	Days
Dry days (DD)	Seasonal total number of dry days (DP < 0.85 mm)	Days
Onset of growing season (OGS)	First occasion from 1 November with 20 mm or more of rain within a 3-day period and no dry spell exceeding 10 days in the following 30 days	Date
End of growing season (EGS)	Last day before May 30 that accumulated 10 mm or more rainfall	Date
Duration indices
Wet spell/pentad (WP)	Five consecutive days with more than 4.25 mm of rainfall	Number
Dry spell (DS)	Five consecutive days with less than 4.25 mm of rainfall	Number
Growing season length (GSL)	Date of the beginning to the date of ending of the growing season	Days
Absolute indices
Maximum 1-day rainfall (Rx1day)	Seasonal maximum precipitation in 1 day	mm
Maximum 5-day rainfall (Rx5day)	Seasonal maximum precipitation in five consecutive days	mm
Percentile-based indices
Extreme intensity (RXI)	Average intensity of events greater than or equal to the 1961–2007 mean 95th percentile of the wet days	mm
Extreme frequency (RXF)	Number of days in the season with rainfall exceeding the 1961–2007 mean 95th percentile of the wet days	Days
Extreme percent (RXP)	Proportion of total seasonal rainfall from all events above the average long-term (1961–2007) mean 95th percentile of the wet days	%
Other indices
Seasonal rainfall (PCPTOT)	Seasonal total precipitation from wet days (DP ≥ 0.85 mm)	Mm

align with standards on climate change detection stipulated by the World Climate Research Programme on Climate Variability and Predictability’s Working Group on Climate Change Detection [33,34]. These indices reflect recognised elements of global climate change, like variations in event severity, frequency, and duration, and are widely recommended. It is also suggested to use indices that capture frequent seasonal or yearly events for stronger statistical reliability and to consider extreme measures that might not occur annually.

The calculation of wet and dry days was based on the total length of the growing season, rather than the difference, to eliminate the confounding effects of zero cases [35]. Contrary to other studies [36], a day with no precipitation was not considered a dry day. A wet day was defined as having a minimum of 0.85 mm of rain, with the inverse definition applying to a dry day [35]. A wet spell, or pentad, was characterised by five consecutive days receiving a total of 4.25 mm of rainfall, based on the minimum threshold for a wet day of 0.85 mm of rain and the corresponding inverse for a dry day or spell [35]. The total number and duration of dry spells were determined and classified into four categories: short (5 days), medium (6 to 10 days), long (11 to 15 days), and very long (more than 15 days). The length of the growing season was calculated by subtracting the start date from the end date of the season. The season was considered to begin on the first day of November—modified from the start date of 15 November used in [37]—with 20 mm or more of rainfall over three days and no subsequent dry spell lasting more than 10 days within the next 30 days. The end of the season was marked by the last day before May 30 that received 10 mm or more of rainfall.

3.2. Statistical and Trend Analysis

Moving average statistics, such as means and the coefficient of variation (CV), combined with statistical methodologies, provide powerful and comprehensive insights for analysis, including the ability to detect non-linear trends [38,39]. The CV categorises rainfall variability into low (CV < 0.20), moderate (CV = 0.20–0.30), high (CV = 0.30–0.50), or extreme (CV > 0.50) [40]. Trend analysis using the Mann–Kendall non-parametric rank correlation test (M-K test) identifies long-term changes in rainfall and dry spell events and describes whether such trends are statistically significant [30,41]. The test’s significance level indicates its strength, while the slope magnitude estimate reveals the trend’s direction and size. For local significance, confidence levels at 90%, 95%, and 99% represent possible, strong, and very strong evidence against the null hypothesis, respectively.

The Mann–Kendall (M-K) test is notable for its low sensitivity to abrupt breaks in inhomogeneous time series and its robustness against outliers, as well as censored and missing values [42]. In this study, all stations showed homogeneous rainfall data, with the number of runs falling within the 5% significance level, as determined by a homogeneity test based on cumulative deviations from the mean [43]. The M-K test assumes measurement independence, meaning there should be no serial dependence between measurements at different times. This serial dependence, or autocorrelation, typically seen in daily data as opposed to monthly data spaced a year apart, can increase the likelihood of falsely detecting significant trends or missing existing ones. Recognising this, many studies have underscored the importance of estimating and adjusting for the autocorrelation coefficient in the sample data before applying the M-K test [44,45,46,47]. This study used the modified M-K test from XLSTAT 2012 software, which accounts for autocorrelation and seasonal effects [48]. Furthermore, it analysed daily rainfall data specifically from the wet season (November to May), avoiding the calendar year approach that could introduce seasonality elements. Lastly, the five time series have previously been validated in studies where tests for randomness and independence revealed no significant serial or spatial correlation [6,28,49].

The study used Cramer’s test to analyse the stability of long-term precipitation records by comparing the overall mean with the means of specific periods [13,50]. This method gauges total precipitation variability beyond just the total amounts. To address this variability, the study established a series of area-weighted normalised precipitation anomalies, which are anomalies scaled according to each station’s average precipitation. Cramer’s test also allowed for the comparison of seasonal means and decadal averages against the overall mean for the entire record period, helping to identify seasonal anomalies and significant rainfall trends within specific decades. The decades were divided into non-overlapping intervals: 1961–1970, 1971–1980, 1981–1990, 1991–2000, and 2001–2007.

For each station and across the years studied, the mean ($\overline{\mathrm{x}}$) and the standard deviation (δ) were calculated. The t_k statistic [13,50], representing a moving t-statistic, measured the difference between the decadal mean ($\overline{\mathrm{x}}$k) and the overall mean ($\overline{\mathrm{x}}$). The resulting values were then evaluated against Student’s t-distribution with (N − 2) degrees of freedom, at a 95% confidence level, which is suitable for a two-tailed test. A t_k value exceeding the bounds of the two-tailed probability of the Gaussian distribution (equal to 2.021 at a 95% confidence level for 45 degrees of freedom) indicated a significant departure from the long-term mean.

The relationship’s strength and direction between the co-occurrence of El Niño and La Niña (ENSO) events and extreme rainfall and dry spell events were determined using linear regression. El Niño was associated with 17 seasons over the period, while La Niña and neutral years were each associated with 15 seasons. The Oceanic Niño Index (ONI)—a 3-month running mean of ERSST.v4 sea surface temperature anomalies in the Niño-3.4 region, based on centred 30-year base periods updated every five years—was used to indicate ENSO events [51]. The interpretation of a normality test using the Shapiro–Wilk test showed that the ONI values followed a normal distribution. A non-parametric Spearman’s-Rho rank test was also applied to each time series. This test was designed with the understanding that long-term patterns in extreme events might not be linear and only visible during certain years. The results of these analyses are presented for inferences.

4. Results

4.1. Variability and Trend Analysis of Rainfall and Dry Spell Events

4.1.1. Wet Events

Long-term variability in seasonal rainfall totals (PRCPTOTs), wet days (WDs), and extreme percent (RXP) ranged from low (CV = 0.2) to moderate (CV = 0.2–0.3), as shown in

Table 3. Mean and variation of seasonal (November–May) rainfall and dry spell events for 1961–2007.

Rainfall Indices (Seasonal)	Balaka			Dedza			Bunda			Chitedze			KIA
	Mean	SE	CV	Mean	SE	CV	Mean	SE	CV	Mean	SE	CV	Mean	SE	CV
Seasonal rainfall (mm)	795.3	33.64	0.29	878.0	23.05	0.18	872.0	30.53	0.24	835.1	26.80	0.22	793.6	24.31	0.21
Number of wet days	49.0	2.07	0.29	68.0	1.69	0.17	59.0	1.98	0.23	62.0	1.72	0.19	60.0	1.31	0.15
Number of wet pentads	2.2	0.07	0.23	4.7	0.16	0.24	2.7	0.09	0.24	3.3	0.26	0.54	3.0	0.10	0.22
Max 1-day rainfall (mm)	72.1	3.37	0.32	69.9	3.16	0.31	64.2	2.81	0.30	71.5	2.92	0.28	66.0	3.37	0.35
Max 5-day rainfall (mm)	104.0	9.40	0.62	118.5	7.26	0.42	91.0	7.03	0.53	106.4	5.43	0.35	108.8	7.14	0.45
Extreme intensity (mm)	48.2	2.04	0.29	43.4	1.20	0.19	44.2	1.29	0.20	47.7	1.32	0.19	43.3	1.33	0.21
Extreme frequency (days)	2.8	0.13	0.32	2.7	0.13	0.33	2.9	0.15	0.36	2.6	0.11	0.30	2.9	0.13	0.30
Extreme percent (%)	43.1	1.19	0.19	34.4	0.70	0.14	35.8	1.31	0.25	35.4	1.03	0.20	38.2	0.72	0.13
Number of dry days	82.0	2.75	0.23	59.0	2.24	0.26	75.0	2.63	0.24	63.0	2.66	0.29	61.0	2.85	0.32
5 days dry spells a	1.6	0.12	0.53	1.8	0.13	0.48	2.0	0.14	0.47	1.5	0.09	0.39	1.8	0.13	0.48
6 to 10 days dry spells	7.4	0.18	0.17	7.4	0.19	0.18	7.6	0.20	0.18	7.7	0.21	0.19	7.6	0.19	0.17
11 to 15 days dry spells	12.6	0.18	0.10	12.4	0.18	0.10	12.7	0.24	0.13	12.8	0.21	0.11	12.6	0.18	0.10
>15 days dry spells	20.1	0.67	0.23	18.2	0.45	0.17	20.2	0.53	0.18	17.8	0.34	0.13	19.4	0.57	0.20

^a Data for 5-day dry spells is based on the mean number of dry spell events, unlike other dry spell categories that are based on mean lengths.

. The maximum 5-day rainfall (Rx5day) exhibited the highest variability (CV = 0.35–0.62), followed by maximum 1-day rainfall (Rx1day, CV = 0.28–0.35) and extreme frequency (RXF, CV = 0.30–0.32), both of which demonstrated moderate to high variability. Balaka recorded the fewest wet days and pentads but had the highest values for Rx1day, extreme intensity (RXI), and extreme percent (RXP). Dedza featured the highest readings for wet days, wet pentads and Rx5day.

Linear regression analysis was performed to investigate the influence of extreme rainfall events equal to or above the 95th percentile on a long-term seasonal rainfall regime. The linear fit and statistical significance of the relationship between RXI and RXP with seasonal rainfall are shown in Figure 2 and Figure 3, respectively. RXI exhibited a significant correlation with rainfall across locations, with the highest R² value at KIA (R² = 0.68 ***) and the lowest at Chitedze (R² = 0.34 ***). RXP did not show a significant correlation with seasonal rainfall at Dedza (R² = 0.013), while strong correlations were observed at Bunda (R² = 0.38 ***) and Balaka (R² = 0.29 ***). At all locations, RXF (data not plotted) was not significantly correlated with mean seasonal rainfall (adjusted R² = −0.022 for Balaka, −0.014 for Dedza, −0.02 for Bunda, −0.01 for Chitedze, and −0.009 for KIA).

The Mann–Kendall test revealed significant trends in PRCPTOT, wet days, and the three extreme rainfall indices (RXF, RXI, and RXP) at various sites (

Table 4. Seasonal (November–May) Mann–Kendall trend analysis for the period 1961–2007.

Rainfall and Dry Spell Indices	Kendall’s Tau
	Balaka	Dedza	Bunda	Chitedze	KIA
Seasonal rainfall (PRCPTOT, mm)	0.065 **	0.015	−0.188 ***	−0.010	0.057
Wet days (WD)	−0.225 ***	−0.144 **	−0.188 ***	−0.214 ***	−0.059
Wet pentads (WP)	−0.207	−0.138	−0.001	−0.129	0.173
Max 1-day rainfall (Rx1day, mm)	0.174	0.134	−0.070	−0.029	0.077
Max 5-day rainfall (Rx5day, mm)	0.029	0.111	−0.040	0.040	0.150
Extreme intensity (RXI, mm)	0.157	0.031	−0.050	0.188 ***	0.064
Extreme frequency (RXF, days)	0.054	0.308 **	−0.070	0.100	0.020
Extreme percent (RXP, %)	0.119	0.192	0.231 **	−0.181 ***	0.008
Dry days	0.000	0.078	−0.028	−0.168 ***	−0.094
5-day dry spells (5 DS)	0.012	0.158	−0.093	0.146	−0.025
6 to 10 days dry spells (6–10 DS)	−0.038	−0.228 **	0.013	0.017	−0.149
11to15 days of dry spells (11–15 DS)	0.063	−0.218	−0.025	−0.122	−0.360 **
>15 days of dry spells (>15 DS)	0.161	0.068	0.131	−0.210 *	0.129
Start of the season (days)	0.114	0.007	0.130	0.157	0.054
End of season (days)	−0.108	−0.216 **	−0.076	−0.279 **	−0.148
Growing season length (GSL, days)	−0.197 ***	−0.178 **	−0.187 ***	−0.338 ***	−0.117 **

Kendall tau values: * significant at α = 0.10, ** significant at α = 0.05, *** significant at α ≤ 0.01.

). Seasonal rainfall in Balaka showed a significant positive trend and a significant negative trend in Bunda. Except for KIA, the trends for wet days were significant and negative. At Chitedze, RXI exhibited a significant upward trend, whereas RXF showed a significant negative trend. Extreme percent (RXP) showed a significant positive trend at Bunda and a negative trend in Chitedze.

4.1.2. Dry Events

The spatial pattern and magnitude of the average lengths of dry spells across different classes were consistent across locations, as shown in Table 3. The frequency of 5-day dry spells showed high to extreme variability (CV = 0.39–0.53), while the variability was lowest for the average lengths of 11- to 15-day dry spells (CV = 0.10–0.13). Table 4 indicates that each key dryness index was confined to a single location, and all displayed downward trends. The trends for dry days and dry spells longer than 15 days were significant at Chitedze, whereas the 6- to 10-day and 11- to 15-day dry spells showed significant trends at Dedza and KIA, respectively.

4.1.3. Onset, End and Length of the Growing Season

The length of the growing season (GSL) varied from 129 to 137 days, as detailed in

Table 5. Characteristics of length, start, and end of the growing season (November–May).

Seasonal Characteristics a	Balaka	Dedza	Bunda	Chitedze	KIA
Growing season length (days)	134 (0.17)	136 (0.17)	137 (0.16)	134 (0.17)	129 (0.17)
Start of the growing season
Earliest date	01 Nov	05 Nov	01 Nov	2 Nov	26 Oct
Median start date	23 Nov (15)	27 Nov (14)	22 Nov (15)	25 Nov (15)	24 Nov (15)
Most delayed date	1 Jan	03 Jan	01 Jan	11 Jan	30 Dec
End of the growing season
Earliest date	23 Feb	03 Mar	21 Feb	27 Feb	04 Mar
Median end date	14 Apr (19)	15 Apr (21)	09 Apr (21)	10 Apr (20)	07 Apr (20)
Most delayed date	26 May	27 May	24 May	27 May	25 May

^a Coefficient of variation (CV) for GSL is given in brackets. Standard deviation (days) for the median start and end of the season is given in brackets.

. The median onset and cessation dates of the growing season were remarkably close, occurring within a single pentad of each other, specifically from 22 to 27 November and 7 to 17 April, respectively. The standard deviation (SD) for the median start date of the season was 15 days and remained consistent, while the SD for the median end date of the season varied from 19 to 21 days across different sites. According to Table 4, long-term trends for the onset of the season were positive, indicating a later start, but these were not statistically significant at any site. Except for Dedza and Chitedze, trends for the season’s end dates were not significant but uniformly negative, suggesting an earlier end. The length of the growing season displayed significant and negative trends, indicating a decline across all sites.

4.2. Seasonal Rainfall Departures

Inter-seasonal rainfall departures obtained using Cramer’s trend test are shown in Figure 4. Among the total instances of above-normal departures at Balaka, Dedza, Bunda, Chitedze, and KIA, 38.1%, 52.6%, 63.6%, 50%, and 57.1%, respectively, were found to be significant (t_k = 2.021 (45_df), α = 0.05). In contrast, below-normal deviations were significant in 32%, 41.7%, 52.2%, 37.5%, and 53.8% of cases for the respective series shown in Figure 4; Balaka exhibited the fewest overall departures. There were clearly defined periods of above- and below-normal rainfall, as seen by the 5-year running mean of rainfall departures for Balaka, Chitedze, and KIA. While KIA leaned near the mean, Chitedze had noticeably higher deviation values. Rainfall departures for the Dedza and Bunda series showed synchronous variations, with one above-normal period followed by lower departures. Multi-decadal periods of above-normal deviations were evident in the Dedza series between 1976 and 1993 and in the Bunda series between 1971 and 1985.

4.3. Decadal Epochs of Rainfall

Table 6. Non-overlapping decadal means and t_k values of rainfall series.

Period	Balaka		Dedza		Bunda		Chitedze		KIA
	Mean	tk	Mean	tk	Mean	tk	Mean	tk	Mean	tk
1961–1970	746.4	−0.72	818.5	−1.24	894.8	0.38	813.9	−0.41	762.5	−0.64
1971–1980	817.5	0.33	895.4	0.38	1018.3	2.02 **	870.5	0.67	832.7	0.80
1981–1990	820.6	0.38	975.0	1.84 *	877.0	0.08	866.8	0.60	779.3	−0.30
1991–2000	790.6	−0.07	827.3	−1.08	794.2	−1.23	791.8	−0.81	792.0	−0.03
2001–2007	804.2	0.10	872.4	−0.10	734.5	−1.55	831.1	−0.06	804.8	0.19

* t_k value significant at α = 0.10; ** t_k value significant at α = 0.05.

shows the findings of trend detection in decadal epochs of rainfall using Cramer’s Test. Except for the 1971–1980 period at Bunda and the 1981–1990 period at Dedza, differences between decadal means and long-term averages were not significant. Negative trends were, however, evident for the 1961–1970 and 1991–2000 epochs, followed by two successive decades of above-average rainfall from 1971–1980 to 1981–1990. Generally, the trends were below average in the subsequent two decades.

4.4. ENSO-Induced Rainfall Anomalies

Relationship between ENSO Events and Extreme Rainfall Events

The bivariate relationship between ENSO events, based on Oceanic Niño Index (ONI) values, and extreme rainfall events was determined through Spearman’s rank correlation and linear regression analyses. As shown in

Table 7. Relationship between extreme wet and dry events against the Oceanic Niño Index values for the period 1961 to 2007 based on Spearman’s rank correlation test (DS = dry spell).

Indices	Spearman’s Rho
	KIA	Chitedze	Bunda	Dedza	Balaka
RX1day	0.22 *	0.10 ns	0.08 ns	−0.04 ns	0.11 ns
RX5day	0.28 **	0.02 ns	0.09 ns	0.04 ns	−0.03 ns
RXF	0.13 ns	−0.18 ns	−0.06 ns	−0.10 ns	0.15 ns
RXI	0.23 *	0.07 ns	0.02 ns	−0.10 ns	0.01 ns
RXP	0.20 *	−0.12 ns	0.10 ns	−0.12 ns	0.02 ns
5 days DS	0.07 ns	0.06 ns	−0.08 ns	−0.17 ns	−0.16 ns
6–10 days DS	−0.01 ns	−0.05 ns	−0.13 ns	0.18 ns	−0.12 ns
11–15 days DS	−0.02 ns	0.00 ns	0.07 ns	0.01 ns	0.01 ns
>15 days DS	−0.12 ns	−0.15 ns	−0.03 ns	−0.14 ns	0.04 ns

*ρ value significant at α = 0.10; **ρ value significant at α = 0.05; ns = ρ not significant.

, the results indicate that, with the exception of KIA, Spearman’s correlation coefficients were predominantly insignificant, suggesting a lack of a robust relationship between the co-occurrence of extreme events and ENSO cycles. However, at KIA, the relationship was strongly significant (p ≤ 0.05) for Rx5day and possibly significant (p ≤ 0.10) for Rx1day (p = 0.07), RXI (p = 0.06), and RXP (p = 0.09).

In

Table 8. Linear regression relationship between extreme wet and dry events against the Oceanic Niño Index values for the period 1961 to 2007.

Index and Statistic		KIA	Chitedze	Bunda	Dedza	Balaka
RX1day	Y	4.55x + 65.80	−0.67x + 71.55	6.36x + 63.69	−0.62x + 69.88	0.98x + 72.07
	R2	0.0275	0.0008	0.0776	0.0006	0.0013
RX5day	Y	11.66x + 108.38	−0.20x + 106.41	4.35x + 90.87	0.91x + 118.43	−6.08x + 104.18
	R2	0.0407	0.00002	0.0058	0.0003	0.0064
RXF	Y	0.21x + 2.78	−0.12x + 2.43	−0.11x + 2.94	−0.17x + 2.60	0.25x + 2.71
	R2	0.0089	0.004	0.002	0.0075	0.0124
RXI	Y	2.05x + 43.18	0.92x + 47.65	1.05x + 44.12	−0.99x + 43.48	−0.63x + 48.24
	R2	0.0367	0.0076	0.0099	0.0107	0.0014
RXP	Y	1.05x + 38.18	−1.15x + 35.45	1.36x + 36.39	−0.72x + 34.47	−0.15x + 43.13
	R2	0.0332	0.0181	0.0246	0.0169	0.0003
5 DS	Y	0.50x +6.36	0.09x + 6.17	−0.61x + 6.51	−1.18x + 6.43	−1.11x + 6.53
	R2	0.0063	0.0004	0.0073	0.0325	0.0356
6–10 DS	Y	−1.27x + 17.41	−0.79x + 19.33	−2.52x + 21.41	2.44x + 16.46	−1.84x + 23.41
	R2	0.007	0.004	0.0235	0.0288	0.011
11–15 DS	Y	−1.24x + 3.81	−0.69x + 8.00	1.48x + 8.63	0.17x + 6.40	−0.78x + 11.38
	R2	0.0323	0.0031	0.013	0.0004	0.0033
>15 DS	Y	−1.33x + 7.11	−2.08x + 4.32	−0.95x + 6.74	−1.62x + 3.55	1.48x + 10.67
	R2	0.0134	0.0419	0.0056	0.0347	0.0135

, a negative coefficient in the linear equation (Y) indicates a stronger association between extreme events and dry ENSO occurrences (El Niño), while a positive coefficient indicates a stronger association with wet ENSO events (La Niña). The coefficient of determination (R²) did not show a significant linear association between dry and wet ENSO episodes and extreme rainfall and dry spell events. The non-linear Spearman correlation test appears to capture more complex patterns in the data, allowing for the detection of significant relationships, as seen in Table 7.

5. Discussion

5.1. Seasonal Rainfall

Total seasonal rainfall (PRCPTOT) exhibited a significant negative trend in Bunda and a significant positive trend in Balaka, while non-significant trends were observed in Dedza, Chitedze, and KIA. Conflicting results emerge from various studies, some projecting an abrupt negative shift in the long-term rainfall behaviour, whereas others indicate a general increase in rainfall during the maize-growing season in Malawi and the southern African region [7,52]. Our findings were further compared with several studies that analysed seasonal rainfall trends across Malawi with comprehensive coverage [6,18,53]. In such studies, many locations reported insignificant negative or positive trends for seasonal and/or annual values.

This study revealed significant negative trends in wet days across all locations except for KIA, indicating that these trends did not align with those of total seasonal rainfall (PRCPTOT). This suggests the influence of other rainfall characteristics on seasonal totals. Earlier research noted similar negative trends in wet days for Chitedze [35] and Balaka [49]. Additionally, broader regional studies forecasted a general decrease in both the number of wet days (with at least 0.2 mm of rain) in southern Africa [54] and the frequency of rainy days across Malawi [21].

5.2. Extreme Rainfall Events

Clay et al. [55] highlighted that the emphasis on drought and extremely low rainfall often overshadows the significance of extremely high rainfall events and the upper distribution of rainfall. In many southern African regions, it is argued that climate impacts do not need to reach extremes to significantly affect environmental processes and economic activities, such as agriculture. Since agriculture relies on just enough rainfall, any deviation from the norm can lead to severe consequences [56]. To analyse trends in the intensity of continuous rainfall episodes, the Rx1day and Rx5day indices were used. However, neither index showed significant long-term trends aligning with findings from other Malawi [18,29,35] and southern Africa studies [57].

Three extreme rainfall indices—extreme intensity (RXI), extreme frequency (RXF), and extreme percentage (RXP)—typically capture the top three to five rainfall events annually, accounting for about 20–25% of total rainfall [20]. The variability coefficient for RXI and RXP was low to moderate, while it was high for RXF, indicating that RXF is not a reliable indicator. Linear regression analysis showed that RXP and RXI consistently contributed to the long-term rainfall regime across locations, whereas RXF’s contribution was not significant. Additionally, RXP’s contribution appeared to be as critical in the drought-prone area of Balaka as it was in the rain-favoured mid-altitude plains of central Malawi. Mann–Kendall trend analysis did not show consistent altitudinal or zonal patterns across the indices. However, there were significant isolated trends: RXF showed a declining trend at Dedza, RXP decreased at Chitedze, RXP increased at Bunda, and RXI increased at Chitedze. These findings somewhat contradict the general expectation under global warming scenarios, which predict an increase in the frequency (RXF) of extreme precipitation events [39]. Instead, our results indicate varied intensities and proportions of extreme rainfall even within what might be considered homogeneous climatic zones [8,54].

Despite the lack of consistent long-term trends in these indices, their role in the variability of seasonal rainfall is still crucial. In Malawi, short-duration heavy convective rainfall, covering a few hundred square kilometres, frequently contributes significantly to the duration of wet periods, total precipitation, and the occurrence of floods. The use of extreme indices is also economically justified when introducing, for example, rainfall-indexed insurance to cushion farmers from the vagaries of weather or biophysical interventions in agricultural water management [11,25]. Evaluating the lower and upper tails of the daily rainfall probability distribution at the precise location where interventions are most likely to be undertaken remains prudent. The requirement to inform these and other applications calls for site-specific studies into changes in identifiable key rainfall metrics.

5.3. Dry Days and Spells

Seasonal rainfall is influenced by the occurrence of dry days and spells, which are especially critical to agriculture; crops can be devastated by a hot, dry period, even if seasonal rainfall totals appear favourable. The CV for dry days and spells was low to moderate across locations. With the exception of Chitedze, which displayed significant negative trends, none of the remaining locations showed significant long-term trends in dry days. This contrasts with earlier studies that reported an increasing trend in dry days in some central and southern Malawi locations [7,33].

Regarding dry spells, significant negative trends were identified for spells lasting 6 to 10, 11 to 15, and over 15 days at Dedza, KIA, and Chitedze, respectively. Earlier work found no significant trends in the occurrence of dry spells of 7, 10 and 15-day durations in ten homogenous regions in Malawi [58]. Recent studies have, however, reported an increase in the average length of dry spells during the rainy season in Malawi [21,59]. This was attributed not necessarily to changes in rainfall amounts but to shifts in rainfall patterns. Barron [60] observed that dry spells lasting 14 to 28 days are common in east and southern Africa, potentially affecting crop growing days by 30–40 days and introducing significant uncertainty into farming and water resource management. Furthermore, these dry spells are generally independent of total seasonal rainfall, indicating that forecasts of above- or below-normal rainfall may not significantly assist small-scale farmers who are averse to using such information [34].

5.4. Seasonal and Decadal Rainfall Departures

Variations in the frequency of below-normal or above-normal rainfall have been used to identify climatic variability, which is considered high over southern Africa and has been a consistent feature of the region’s climate for many years [61]. The strong interannual fluctuations in precipitation, characterised by recurring wet and dry periods, are primarily attributed to the influence of coupled ocean-atmosphere modes of variability across the Pacific, Indian, and Atlantic Oceans [62]. In our study, while all stations exhibited interannual variability, Bunda and KIA demonstrated a higher proportion of deviations. In contrast, Balaka, known to be drought-prone, showed the fewest departures. Notably, the trend for Balaka included periods with significant peaks on a cycle of approximately 5 to 6.3 years, aligning with the patterns described by Nicholson et al. [8]. After 1991, Bunda and Dedza experienced below-average anomalies, consistent with the observations made by Mwafulirwa [3].

The time series analysis indicated no significant decadal rainfall variability for most locations, except for positive trends in the 1971–1980 and 1981–1990 decadal means for Bunda and Dedza, respectively. Tadeyo et al. [52] observed similar patterns, with positive decadal changes in the 1980s, followed by a significant decline from 1999 to 2008. Negative trends were evident for the study locations for the periods 1961 to 1970 and after 1990, which corresponds to observed drying patterns across southeastern Africa and changes in the water balance components [63,64]. The consecutive positive or negative trends highlight the role of multi-decadal variability in Malawi’s climate. Hurrell et al. [65] pointed out that although decadal climatic changes can be evident, they typically are overshadowed by the internal variability stemming from the slow-changing components of the climate system in different areas. Understanding these inter-seasonal and decadal trends is crucial for developing strategic policies, decisions, and investments to adapt to climate variability, which may require several years to implement.

5.5. ENSO-Induced Rainfall Anomalies

Studies have demonstrated a strong link between the ENSO phenomena and significant anomalies in rainfall and dry spells, substantially affecting extreme wet and dry events [9,66]. In our study, except for KIA, the findings showed statistically insignificant correlation coefficients, indicating no significant relationship between the occurrences of wet or dry extremes and ENSO cycles. Further examination of the relationship between ENSO cycles and dry spells revealed non-significant yet higher instances of dry spells during ENSO cycles in both moderate (Dedza) and vulnerable (Balaka) locations, with fewer instances of dry spells coinciding with wet ENSO cycles across the sites. These findings are consistent with the long-term trends for dry days, which indicated a positive, though non-significant, trend.

Malawi is described as being situated between the eastern equatorial and southeastern ENSO regions, exhibiting conflicting climate responses that complicate the determination of inter-seasonal ENSO impacts on rainfall [9]. Under El Niño conditions, the former region typically receives above-average rainfall, while the latter usually experiences below-average rainfall, with a contrasting pattern observed during La Niña episodes [5]. The extent of the area affected varies with each El Niño or La Niña event, with pervasive effects in the north of the country and mixed responses in different agroclimatic zones throughout Malawi [9]. This variability underscores the importance of cautious interpretation in studies and the application of climatic regionalisation that fails to account for the characteristics of local observational sites, as emphasised in our study.

The relationship between the strength of trends in extreme rainfall, dry spells, and the co-occurrence of ENSO events has been extensively evaluated and debated. Pomposi et al. [67] argued that, despite a known connection between El Niño events and precipitation in southern Africa, not all El Niño events have led to drying in the region. Kane [68] observed the occurrence of severe droughts both before and after El Niño events and suggested that other physical processes might be responsible for extreme dry events in the area. For example, drought-induced anomalies caused by El Niño events were found to be more closely related to changes in temperature than to rainfall [69]. Therefore, it is crucial to consider a variety of factors, including temperature, in studies examining the impact of ENSO-driven anomalies on crop cultivation [17,70,71].

5.6. Onset, Cessation and Length of Growing Season

Several studies in Malawi have reported evidence of a shortening growing season, while others have found no significant changes in length [7,8,21,47,72,73,74]. For instance, projections by Vizy et al. [75] suggest a potential shortening of the growing season by the mid-century in Malawi, with reductions of 20–55 days in the southernmost districts and 5–30 days in the central districts. This anticipated shorter growing season is primarily due to an earlier end date, as no significant change in the season’s start is predicted. Our study reveals a trend towards later starts of the growing season that, although not statistically significant across all locations, contrasts with significant trends in earlier ends of the season, notably in Dedza and Chitedze. This pattern aligns with a significant decrease in the overall length of the growing season across the studied locations.

The criteria for determining the onset, end, and length of the growing season vary across studies. Our study adopts the guidelines developed by Stern and Cooper [35] and those of Stroosnijder and Hoogmoed [76], which categorise rainfall events at the season’s onset into less significant (<10 mm), common (10–20 mm), and rare but impactful (>20 mm) events. The Malawi Department of Climate Change and Meteorological Services (DCCMS) defines the season’s start as receiving at least 40 mm of rainfall within 10 days, provided there are no dry spells of 10 or more consecutive days within the month. The season’s end is identified when rainfall does not exceed 25 mm in 15 days starting from 15th March. Conversely, other studies define the onset as 25 mm of rainfall within 10 days without a subsequent 10-day dry period and the end as three consecutive periods of less than 20 mm of rainfall each after February 1 [72]. The variation in thresholds for determining the start and end of the rainy season could impact the accuracy of historical studies or the forecasting of seasonal rainfall.

Using pentad analysis, research showed that Malawi typically experiences its first major wet spell from 2 to 6 December, give or take two pentads [3]. Recent studies have indicated that the onset of the rainy season across Malawi is relatively uniform, occurring from mid-November to early December, with the season generally ending from mid-March to early April [8]. Our study found that the median date of onset across all stations was within one pentad, 22–27 November, with a standard deviation of 15 days, aligning with most existing research. Thus, a typical 125-day maize variety planted on 22 November would ideally mature in the pentad before the season’s median finish date, which was determined across locations within the pentad of 7–14 April. The other stipulations cited for the end of the growing season posit potential for a false end date for 125-day crop duration.

Crop evapotranspiration indices have been used to define the onset, cessation, and duration of the growing season. Fiwa [47], working in central Malawi, including the current study areas of Bunda and Chitedze, identified a trend towards a delayed start and reported mean end-of-season dates between April 18 and 22, which is about seven days later than what our current study finds. However, employing crop indices effectively requires extensive and accurate data on crops, soil, and climate, along with the necessary expertise. Despite this, analysing additional variables that combine precipitation with temperature, evapotranspiration, and soil moisture could enhance the use of rainfall indices, contributing to more comprehensive hydroclimatic assessments for developing interventions.

5.7. Location Inferences

This study recognises that the variability in the number and types of indices reflecting long-term trends across sites suggests that not every index will accurately represent spatial and temporal variations in rainfall. Local geographical features and the varied impact of synoptic-scale systems may influence each site differently, potentially leading to certain indices revealing or failing to detect spatial and temporal changes, even within what are considered homogeneous climatic zones. For example, significant differences in the general long-term climatological characteristics are observed among stations in the Lilongwe plain. Notably, KIA, situated north of the Lilongwe Plain, experienced several extreme rainfall events associated with wet ENSO episodes, unlike other locations in the same climatic zone. Balaka recorded the fewest wet days and Dedza the most. However, Balaka showed higher long-term mean values for Rx1day, RXI, and RXP, seemingly offsetting its greater number of dry days, while the highest mean value for Rx5day appeared to enhance Dedza’s rainfall regime. Balaka stands out in discussions on dry events. Despite having more dry days, its variability and trends in dry spells were indistinguishable from those at other sites. This finding contrasts with studies like those by [34], which observed that in southern Africa, areas with low and consistent rainfall tend to experience a higher number of dry spells, whereas regions with more consistent high rainfall tend to have fewer.

Previously, climatological studies have suggested that climatological/geographical regionalisation or averaging recorded station data over larger areas are effective methods for observing climate trends, as the impacts of local variations are minimised [8,18,19,20]. These techniques are acclaimed for helping navigate data heterogeneities and detailed nuances, focusing instead on the key productivity trends influenced by large-scale climatic factors. In agricultural practises, however, the principle of location specificity is paramount for placing biophysical interventions to enhance resource use and crop yield. This is because large-scale climate patterns may not align with the genotype x environment interactions that underpin gains in agricultural productivity. Therefore, changes in rainfall and extremes of dry periods in typical agricultural regions need to be detailed at the local scale to potentially reduce the vulnerability of farmers and their lands to drought and floods. There is an urgent need to invest in data collection, modelling infrastructure, and expertise at the local level to predict potential future impacts of rainfall and dry spell occurrences on agricultural productivity. Development partners and African countries should act within the framework of the United Nations Framework Convention on Climate Change to intensify efforts to improve and broaden the harnessing of and access to climate data.

6. Conclusions

The study determined rainfall indices that establish a baseline for evaluating or planning adaptation strategies at sub-watershed levels based on capturing long-term rainfall variability and changes. From 1961 to 2007, indices were developed to characterise seasonal rainfall, extraordinary rainfall events, and dry spell occurrences aligning with the typical maize growing season in Malawi, from November to April.

Regarding significant long-term changes in total seasonal rainfall, the findings at the five study sites were mixed, showing non-significant trends at three locations and significant positive and negative trends at Balaka and Bunda, respectively. Despite a noticeable reduction in wet days, variations in other rainfall characteristics seemed to offset their impact on long-term seasonal trends. Analysis of extreme rainfall intensity, proportion, and frequency based on the 95th percentile showed no uniform patterns across any indices, with only a few isolated cases of significant declines or increases accompanied by diverse long-term average values. Similarly, dry spells presented a varied pattern, with trend analyses indicating mostly inconsistent and non-significant results.

Variability in long-term interannual rainfall anomalies was noted, with the largest deviations observed in Bunda and KIA, a region with a moderate climate, and the smallest in Balaka, a rainfall-averse zone. Even within the homogeneous Lilongwe plain, data indicated diverse strengths and directions in the correlation between ENSO cycles and episodes of extreme rainfall and dry spells, although largely insignificant. These and proceeding findings support the need for climatological research to extend beyond national or regional variability to include detailed analyses of local geographical areas.

The analysis revealed a consistent onset and finish date, as well as a shortening of the growing season across sites. However, it also found conflicting definitions of the start and end of the rainfall season that determine the length of the growing season. Establishing uniform and objective criteria for determining the season’s onset and cessation is vital for selecting crop varieties with different maturity periods and for current and future sustainable agricultural practises sensitive to shifts in rainfall patterns.