Comparing farmers’ perceptions of climate change with meteorological data: A case study of livestock farmers in Eswatini’s lowveld region

Abstract This paper compared livestock farmers’ perceptions of climate change with meteorological data with the intent of understanding the predictors of farmers’ perceptions. Primary and secondary data sources were used. The sample size was 278 farmers. The data was analysed using descriptive and inferential statistics. The results showed that livestock farmers are mainly aged, married males that practiced mixed farming and reared mainly cattle and goats. The farmers’ perceptions were that the climate had changed notably with increased temperatures, decreased rainfall, delayed rain season and an increased intensity of drought. Climate trends indicated a significant decline in temperatures in the northern lowveld and an increase in the southern lowveld. There was pronounced rainfall variability and a significant decline in summer rainfall in the north. Comparison of climate trends with the farmers’ perceptions revealed some incoherent correlations regarding temperature in the north and annual rainfall throughout the lowveld. Age, education, experience, location, who managed the farm and the crops grown were found to be significant predictors of farmers’ perceptions using multinomial logistic regression. Focusing interventions on these bio-demographic variables will positively enhance farmers’ perception and adaptation. It is further recommended that future studies incorporate drought and floods in comparing trends and perceptions.


Introduction
One of the greatest challenges faced by countries around the world, this century, is climate change. Evidence for climate change is abundant: from the poles in melting ice caps, to the depths of the oceans (increasing sea temperature), to extreme droughts and floods around the world. Dodman and Mitlin (2015) indicate that climate change represents one of the core topics in the worldwide environmental politics agenda. The rationale for this corresponds with indisputable changes or warming to the earth's climate primarily due to human-induced activities (Intergovernmental Panel on Climate Change, 2018). Climate change not only refers to increasing worldwide temperatures due to an increase in green-house gases, but to change in climate conditions or parameters such as temperature, rainfall, wind, humidity, air pressure and severe weather events (Brown et al., 2013). Evidence of anthropogenic impacts on the earth's climate systems reveals that there has been heating up of the planet in the last fifty years mainly due to the discharge of GHG from the combustion of fossil energy sources, forest clearing (Intergovernmental Panel on Climate Change, 2018) and beef production (Hong et al., 2022).
Any change in climate has or will have an impact on people and societies living on earth because such change impacts on their environment that provides for peoples livelihoods, thus such impacts increase people's and societal vulnerability (Niang et al., 2014). Changes in climate affect agriculture either favourably or unfavourably (Gornall et al., 2010) for instance, in several grain producing areas around the world, warming due to climate change has adversely impacted grain production, while in the colder sub-arctic regions warming has positively impacted yields (Porter et al., 2014). Other vulnerabilities to agriculture excluding temperature increases include a reduced growing season and increased risk of drought (Ramirez-Villegas et al., 2021). Besides affecting crops, climate change affects grazing biomass production that affects availability of livestock fodder which affects the livestock health, the livestock growth, fertility, milk and meat production and ultimately resulting in reduction of herd size (Montcho et al., 2022).
To minimize climate change impacts, action plans to minimise the impacts must be developed and implemented (Iglesias et al., 2012). Adaptation strategies or actions could include an array of social, economic, technical, or environmental solutions (Biesbroek et al., 2013). Farmers' adaptation to climate change is related to their perceptions of the changes in the climate (Kuivanen et al., 2015;Mulenga et al., 2017). Therefore a comprehensive understanding of factors persuading farmers views of climate change are essential in understanding adaptation (Aswani et al., 2015;Mulenga et al., 2017).
Usually the adaptation strategies used by farmers are usually, based on the climate changes observed, yet, farmers regularly misconceive climate change which in turn affects adaptation (Asare-Nuamah & Botchway, 2019). Accordingly information of how farmers frame climatic trends and associated climate change implications, can support the development and implementation of improved adaptation strategies (Aswani et al., 2015). Consequently an in-depth understanding of how small-scale farmers in Eswatini contrive climatic trends and their implications to their farming, can be very useful in contributing to minimising the impact of climate change to the agriculture sector. This paper, consequently, studies the livestock farmers' perceptions of climate change, the determinants of these perceptions and compares these perceptions with the meteorological data in order to evaluate how the farmers' perceptions mimic the empirical climatic trends.
The objectives of this study are fourfold: (1) to assess livestock farmers perceptions of climate change in the lowveld region of Eswatini; (2) to assess the trends of climate change in the lowveld region of Eswatini; (3) to determine the predictors of the farmers perceptions of climate change; and (4) lastly, to compare meteorological records with farmers' perceptions regarding climate change.
This study contributes to literature by; firstly, being the first study in Eswatini that determines the perceptions of livestock farmers within the entire lowveld of Eswatini using the mixed methods (qualitative-quantitative) approach. Secondly, previous related studies (Manyatsi et al., 2010;Nkondze et al., 2014) considered localised areas within the lowveld and did not consider other climate indicators such as monthly seasonal changes in climate as well as droughts, floods and winds, this study does consider other climatic variables.

Materials and methods
The Kingdom of Eswatini, situated in Eastern Southern Africa, covering an area of17370km 2 (Manyatsi et al., 2010), has a population of about 1.172 million (World Bank, 2022) of which about 77% reside in rural areas where they practise subsistence rain-fed mixed farming (Thompson, 2016). Eswatini is divided into four physiographic regions namely: the Highveld, Middleveld, Lowveld, and the Lubombo Plateau (Thompson, 2016). The area of interest, the Lowveld (Figure 1) is classified as semi-arid savanna (Sweet & Khumalo, 1994) and is the hottest and driest of the other regions, but is, however, the most productive in terms of livestock production. As a case in point livestock production is the second biggest agricultural industry after sugarcane production in the Lowveld of Eswatini (Thompson, 2016). However droughts and water scarcity in this region are the most common limiting factors to production and these impacts are and will be exacerbated by climate change as illustrated by a study by Tfwala et al. (2020), which showed that the prevalence and severity of droughts in the lowveld is increasing.

Research design and sampling techniques
This study combined information from two sources namely from livestock farmers and climatological data from meteorological stations within the study area. The primary data was obtained from the farmers using a survey questionnaire during late 2020 and early 2021. Information collected consisted of the farmers' bio-demographic information, their climate change knowledge, their This study made use of multi-stage area sampling that involved local constituencies (administrative areas) and farm households, (Kgosikoma et al., 2018). For the local constituencies 10 out of the 14 constituencies were randomly selected for the study in a first stage of sampling, (this equates to a margin of error of 10%, confidence level 70% and sample proportion of 50%) using an online random sample calculator by Raosoft (Raosoft, 2004). At the second stage, 30 farm households from each of the ten constituencies were then randomly selected to give a total of 300 sample households, (of the 300 farmers 278 finally participated which was above the required sample size of 270). The sample size was calculated for farmers' population of 10,776 and with a margin of error used was 5%; the confidence level was 90% and sample proportion of 50% (Raosoft, 2004). The constituencies selected were Mhlangatane, Mandlangempisi, Mkhiweni, Dvokodvweni, Gilgal, Siphofaneni, Mpolonjeni, Sithobela, Somnthongo and Matsanjeni ( Figure 2). As the participants in this research were human subjects, and in order to ensure their rights were protected and proper research ethics were followed, the researchers obtained approval for the research instruments from the UNISA's College of Agriculture and Environmental Sciences Ethics Committee and furthermore, participants consent was obtained before they participated in the research.
The methods for analysis combined descriptive, associational, and inferential statistics. The descriptive data analysis was mainly used on the bio-demographic information gathered from the farmers. While associational and inferential statistics were used mainly on the perception and climate data obtained from the farmers. The main statistical analysis packages employed for the analyses in this study were SPSS Version 20 used for descriptive analysis and multinomial logistic regression analysis and Excel template MAKESENS application for detecting climate trends using the Mann-Kendall test and Sen's slope estimates (Dawood, 2017).

Model specification and justification
In this study, multivariate analysis involved the use of the multinomial logit model to analyse the predictors of climate change perceptions. The review of literature illustrated that the multinomial logit (MNL) model was commonly used by scholars (Deressa et al., 2009;Teye et al., 2015) for studies where respondents had to choose only one option from a certain set of perception or adaptation options. This model was judged to be appropriate because the outcome variables (perception of climate change) had three independent response types, namely: (1) "No change" in climate variables (e.g., temperature, rainfall, etc.) (2) "Increase" in climate variables (e.g., temperature, rainfall, etc.) (3) "Decrease" in climate variables (e.g., temperature, rainfall, etc.) The response "No change" in climate variables was selected as the reference level and was compared with the estimated coefficients of the other two categories "Increase or Decrease" perceived change in climate variable. The climate variable namely: temperature, rainfall amount, rain season and drought frequency were the dependent variables used in the MNL model.
The study considered the following explanatory variables: age, gender, marital status, education, household, size, constituency (area of residence), employment status, occupation, crops farmed, climate change knowledge, access to climate information, farm manager, ownership of cows, ownership of small livestock (goats or sheep), lost livestock to drought, years keeping livestock, livestock farming income and livestock farming expenses.
The MNL model for the perception options specifies the following relationship between the probability of choosing option y i and the set of conditioning variables x as follows: In specifying the MNL model, y i denotes a random variable taking on the values 0, 1, to j, where j is a positive integer, and x i denotes a set of conditioning variables. In this case, y i denotes perception options or categories and x i contains different farmer's characteristics that represent the different personal, household and environmental attributes affecting the farmer's perception options (explanatory variables). β j is a vector of coefficients on each of the independent variables x i . Eq.
(1) can be normalized to remove indeterminancy in the model by assuming that β 0 ¼ 0 and the probabilities can be estimated as:

Biographical results
The results from the 278 livestock farmers interviewed indicated that 78% of the respondents were males and 22% were females. With 66% of the respondents' ≥50 years, 24% aged 30-49 years, and 10% aged ≤29 years. Of these respondents 62% were married; 18% were widowed; 17% were single and only 2% divorced or separated. The majority (74%) live in households having 1-10 persons with only 26% living in larger households. The educational level of the respondents was dominated by farmers who had primary or post primary education (78%) with only by 22% of the farmers having no formal education. All the respondents practiced mixed farming. Furthermore most (51%) of the respondents have been rearing livestock for more than 21 years; 26% have been keeping livestock for 11-20 years; and 23% have been keeping livestock for 0-10 years. Out of the 278 livestock farmers ( Figure 3) interviewed, 254 (91%) raised cattle, 227 (82%) reared goats, 196 (71%) raised chickens, 41 (15%) raised pigs, 10 (4%) farmed sheep, and five (2%) raised donkeys. The average head size was as follows: 29 for cattle, 31 for goats, 32 for sheep, 18 for pigs, 50 for chickens, and seven for donkeys.

Farmers awareness of climate change and access to information
Understanding the farmers' knowledge of climate change is vital in understanding their perceptions of climate change. The results indicated that the majority of farmers (89%) indicated that they understood what climate change was and only 11% did not. Of these, 83% stated they knew the problems associated with climate change, and 17% were not aware ( Figure 4). While 78% of the farmers had access to climate change information and 22% do not. Finally, 76% of the farmers indicated they had access to climate change information relevant to livestock farming.

The sources of climate information for the farmers
The farmers' information sources on climate are ( Figure 5); Radio used about 81% of the time by farmers; Television with a usage of 68%; family and neighbours (with a mean use of 59%); newspapers (with a mean use of 58%); other farmers (with a mean use of 56%); social networks or media (51% mean usage); formal government extension services (51% mean usage); and farmers co-operatives or entities (45% mean usage). Weber (2010) states that mankinds' general perception of climate change and variability is usually based on their observations of climatic variables that affect their lives. For this study, perceived changes of climatic variables over lengthy period of time were used to gauge the farmers' perception. `The results ( Figure 6) indicate that almost all the farmers (99.2%) perceived temperatures and hot days had risen with 0.4% perceiving no change and 0.4% perceiving a decline in temperatures and hot days. For drought (severity and frequency), 89.2% of the farmers perceived an increase in droughts, and only 4.3% perceived no change and 6.5% perceived a decrease in droughts. For wind (severity or frequency), 69.4% of the farmers perceived an increase in wind, 17.3% perceived no change in wind, and 13.3% perceived a decline in winds. Whilst 53.6% farmers perceived a decrease in floods (frequency and severity) 24.1% perceived an increase and only 22.3% perceived no change in floods. Whereas, 49.6% of the farmers perceived that cold days had decreased, 35% of farmers perceived an increase in cold days and 15% perceived no change in  cold days. For the rain season, 71.9% of the farmers perceived a shortened rain season, 22.7% perceived no changed and 5.4% perceived an extended rain season. A total of 91.4% of the farmers perceived a decrease in rain days, 5.4% perceived no change in rain days and 3.2% perceived an increase rain days. Lastly, 92.4% of the farmers perceived that rainfall had decreased, 4% perceived no change in rainfall and 3.6% perceived an increase in rainfall.

Climate trend analysis
The trends were determined using temperature data from two weather stations namely: Big Bend in the south, Mhlume in the north. For rainfall the Sithobela rainfall data, was added to the two weather stations.

Trend analysis of temparature in the lowveld
Trend analysis was conducted for the mean, maximum and minimum temparatures for Mhlume and Big Bend, the detailed results are outlined below.  Figure 8). While, maximum temperatures ranged, generally, between 25.5 o C and 32°C, where, minimum temperatures ranged

Trend analysis of rainfall in the Lowveld
The graphical presentation (Figure 9) of the annual total rainfall for three rainfall measuring stations in the lowveld indicated that there was great variance in rainfall over the years with Mhlume rainfall showing a decreasing trend in total annual rainfall while Big Bend and Sithobela stations showed an increasing trend in rainfall.
On further analysis of the rainfall trends the M-K Test and Sen's Slope test results [ Table 3 near here] indicated that: Even though Mhlume rainfall had a decreasing trend, it was insignificant, the trend was however significant (p ≤ 0.05) for a decrease in the summer rainfall by 186.93 mm over 31 years. Similarly the following months also had a significant decrease in rainfall: February's rainfall decreased (p ≤ 0.05) by 64.33 mm while December's rainfall decreased (p ≤ 0.05) by 67.20 mm over 31 years.
For Big Bend rainfall the only significant change in rainfall was for the month of June that decreased (p ≤ 0.1) by 3.348 mm over 31 years. Similarly for Sithobela the significant exceptions were for the months of June whose rainfall decreased (p ≤ 0.01) by 17.285 mm and October whose rainfall decreased (p ≤ 0.1) by 31.801 mm over 31 years.

Predictors of climate change perceptions
The study made use of the Multinomial Logistic regression (MNL) Model to study the predictors of climate change perceptions. Before running the model, using the SPSS software version 20, eighteen (18) variables assumed to influence perception were checked for the presence of Multicollinearity using Variance Inflation Factor (VIF) indicator where a VIF less than 5 was deemed an acceptable level of multi-collinearity (Daoud, 2017), after this test variables that had acceptable collinearity were used for the Multinomial Logistic (MNL) regression models. [ Table 4 near here] summarizes the Predictor variables used in the models.
The MNL regression results for the predictors of the farmers' perceptions of selected climate variables (temperature, rainfall, rain season and drought frequency) are outlined in [Table 5 near here]. No explanatory variables were significant predictors for perceived changes in temperature. While one variables (age) was a significant (p < 0.001) predictor of perceptions of an increase in rainfall. For age, young farmers (18-39 years) were 8.466 times more likely to perceive "increased rainfall" as opposed to perceiving a "no change" in rainfall. While farmers whose location was the southern Lowveld were 7.869 times more likely to perceive "decreased rainfall" as opposed to a "no change" in rainfall.
Whereas six predictor variables namely education, farming experience, age, types of crops farmed, who the farm manager is and location were significant predictors of perception of a changed rainfall season. The perceptions for an increasing rainfall season were significantly predicted by education (p < 0.1), age (p < 0.05), and farmers location being in the South (p < 0.1). With regards to education, a unit increase in years of schooling reduced the odds of perceiving "an increased rainfall season" by 0.858 times than perceiving a "no change in the rainfall season". While young farmers (18-39 years) were 10.430 times more likely to perceive "increased rainfall season" as opposed to perceiving no change in rainfall season. Likewise farmers who lives in the south are 0.245 times less likely to perceive an increase in rainfall season. A shortening rainfall season was significantly predicted by the number of years keeping livestock or experience (p < 0.05); the farmers age (p < 0.1); the types of crops farmed (p < 0.1) and who managed the livestock farm (p < 0.1). With respect to the farmers' experience, a unit increase in years of experience increased the likelihood of perceiving a shortening rainfall season by 1.046 times, than perceiving a no change in the rainfall season. While younger farmers (18-39 years) were 2.594 times more likely to perceive a "shortening rainfall season" instead of perceiving a "no change in rainfall season", mono-cropping farmers were 1.870 times more likely to perceive a "shortening rainfall season" instead of perceiving a "no change in rainfall season". Whereas household heads who are farm managers are 0.39 times less likely to perceive a "shortening rainfall season" instead of perceiving a "no change in rainfall season".  One predictor variable (constituency or location) was significantly (p < 0.001) associated with farmers' perception of an increase drought frequency. Where farmers whose location was the southern Lowveld were 0.000001 times less likely to perceive 'an increase in drought as opposed to a 'no change in drought.

Comparing farmers perceptions with meteorological data
The farmers' perceptions of climate over an extended period of time (20-30 years) were compared with the trends of climate data obtained from two meteorological stations and one rainfall collecting station within the Lowveld of Eswatini, similar approaches were used in studies by Gbetibouo and Ringler (2009) and Teye et al. (2015). To assist evaluate the farmers' perception to climate, the responses were linked to proximity to the meteorological station.

Comparing temperature perceptions
The farmers' general perception of temperature was that temperatures had increased, (Figure 10). When evaluating their perceptions against the meteorological data, it was observed that mean annual temperatures in the northern Lowveld (using Mhlume data) had declined by 0.5 °C over 31 years with the exception of the month of June whose temperatures have increased by 1.3 ° C over 31 year period. On further examination all monthly maximum temperatures had declined by 1.14 °C, while there was no decline in minimum temperatures.
Whereas actual temperatures in the central and southern lowveld, using the Big Bend data, are in agreement with the farmers' observations. With annual mean temperatures increasing by 0.87° C over the 31 year period while mean winter temperatures increased by 1.13°C . The annual maximum temperatures increased, by 0.87°C over 31 and winter temperature increased by 1.16° C. While annual minimum temperatures increased by 0.88 0 C and an increase in autumn and winter temperatures by 1.48°C and 1.05°C respectively.

Comparing rainfall perceptions
With regard to rainfall, three meteorological stations were used to compare the farmers' perceptions, and these were: Big Bend, Sithobela and Mhlume. The analysis revealed no significant change in Big Bend's and Sithobela's annual rainfall but a significant decrease in June rainfallwith 3.348 mm for Big Bend and 17.285 mm for Sithobela. Furthermore there was a decrease of 31.801 mm in October rainfall at Sithobela. While there was no significant change in Mhlume's annual rainfall, summer season rainfall decreased by 186.93 mm; likewise, the rainfall for the month of February decreased by 67.20 mm over 31 years.
The comparison of the farmers' perceptions of rainfall by region (Figure 11), and the actual rainfall data indicated that there is an overwhelming perception that rainfall decreased, with very few farmers perceiving no change or an increase in rainfall. Although the significant decline does not correlate with annual rainfall, it does with some months or seasons, and this change can influence farmers' perception of rainfall.

Farmers knowledge and sources of climate information
From the descriptive or biographical results it is evident that the farmers' climate change perceptions correlate with their climate knowledge. These results indicated that over 80% of the farmers were knowledgeable of climate change thus indicating a high level of understanding of climate change, similar observations were made by Ajayi (2014) and Ado et al. (2019) who reported awareness of over 84% while studying farmers perceptions in Nigeria and Niger respectively. Likewise, Mandleni and Anim (2011) reported an 85% level of climate knowledge while studying farmers in South Africa. Correspondingly, the use of radio as a climate change information source was comparable with the study by Ajayi (2014) and Diouf et al. (2019) which was used by 81.4% of the farmers, however, differences in use for other information sources such as television, social media or other farmers was significantly more than the studies by Owusu et al. (2021) and Muema et al. (2018). This study revealed that the farmers used a number of sources for climate information dominated by radio, television and newspapers while formal extensions services were less favoured Weber (2010) states that the public's general perception of climate change and variability is usually based on their observations of climatic variables that affect their lives. The results indicated that, in general, the farmers perceived that the following climatic variables were increasing: temperature, hot days, drought severity, drought frequency, and winds; while rainfall, rain days, the rain season, cold days and floods were perceived to decrease. Furthermore, there was spatial variance in how farmers perceived climate change across different locations. These results are similar to those of Maddison (2007), who surveyed farmers in 11 countries across the African continent, and the results concluded that the vast majority of farmers suggested that temperatures had already risen and rainfall had dwindled. Similar findings from sub-Saharan Africa were made with regards to temperature and rainfall, rain-season by scholars such as Debela et al. (2015), Mulenga et al. (2017), andIdrissou et al. (2020). In addition Idrissou et al. (2020) also observed that farmers perceived that winds had increased due to climate change. While studies by Jordaan et al. (2019) and Tfwala et al. (2020) revealed similar findings on droughts.

The climate trends in the lowveld
The temperatures trends for the southern Lowveld were relatively similar to those observed by Archer et al. (2010) who pointed out that the Sub-Saharan region has realized a heating trend over the last decades, while Engelbrecht et al. (2015) and Jury (2018) made similar conclusions about Southern Africa. Similar results were noted by Nkondze et al. (2014) for the central Lowveld. The rise in temperatures, over the 31 years, for the southern Lowveld is significantly higher than the rise in global average surface temperatures for 132 years, where temperature rose by 0.85°C for the period between 1880 and 2012 (Stocker et al., 2013). In the northern Lowveld, the findings deviate from those noted in the South and such could be explained by the undulating topography which can cause many variations to regional temperature, similar findings were reported in the Kingdom of Swaziland's Third National Communication to the UNFCCC (Government of Swaziland, 2016).
With regard to rainfall, the outcomes are comparable with those of Nkondze et al. (2014) and those reported and projected in 2016 Kingdom of Swaziland's Third National Communication Report to the UNFCCC which revealed a great variability in rainfall, while in the long term downscaled projections indicate negative changes and a positive change over the northern part of the country (Government of Swaziland, 2016). These findings also concur with a study on droughts in Eswatini by Jordaan et al. (2019) and Tfwala et al. (2020) who observed that droughts in the lowveld had increased in terms of their incidence, severity and geospatial coverage. The observed variations in rainfall might be related to Southern Africa's current climate drivers, such as the El Niño Southern Oscillation, which is known to negatively affect summer rainfall (Hoell et al., 2017;Funk et al., 2018). Thus, studying the impact of ENSO on rainfall might yield better understanding of rainfall patterns in the Lowveld of Eswatini this is also suggested by Groisman et al. (2005) who proposes that extremes climate events will be used for future climate change analysis.

The predictors of climate change perceptions
The results of MNL regression for exploring the predictors of climate change perception indicated that variables that significantly predicted the farmers' perceptions of climate change parameters included; age, and area of residence (location) which were predictors of an increased rainfall perception and decrease in rainfall amount respectively. Whereas education, farming experience, age, types of crops farmed and the farm manager being head of household, and location were the significant predictors of a change in the rainy season perception. Where education, age, and location being predictors of a lengthening (increased) rain season and farming experience, age, types of crops farmed and the farm manager being head of household, being predictors of a shortening rain season. However farmers' location was the predictor of both an increase in drought. These findings are similar to those of cattle farmers from Benin done by Idrissou et al. (2020) that also revealed that factors determining farmers perception of climate change were cattle farmers experience, the farmers education and the household size, the difference was for, membership of a farmers' association and herd size which were not included in this study analysis. Similar findings were observed by Debela et al. (2015) with regard to age and education, Debela et al. (2015) also found that education and age influenced farmers' rainfall and rain season perceptions, where the older and more literate farmers were inclined to perceive rainfall accurately as opposed to younger and less literate farmers. From this, it is evident that age of the farmer, which can also be a proxy for experience, and education increase the farmers' knowledge of climate change. Similarly the farmer's area of residence or location was one of the most common predictors of perception categories and its influence varied within the region. A similar observation was made by Habtemariam et al. (2016) on farmers in Ethiopia, where location was related to many of their perception categories. The main reason is that location and experiences about climate influence the farmers' perceptions about climate change such findings are comparable with those of Haq and Jafor Ahmed (2020) who observed that people's familiarities with their areas of origin weather events affects their climate change perceptions.

Comparison of farmers climate perceptions with actual data
The farmers perceptions on temperature and rainfall were compared with the climate trend data and the results of this comparison were not fully coherent with the farmers perceptions, for instance, temperature in the northern Lowveld declined, but had increased in the southern lowveld. While for rainfall there was a significant reduction in summer rainfall in the northern lowveld while there was no significant change in the annual and seasonal rainfall in the southern and central lowveld. These findings for the northern lowveld are not similar to those of Mulenga et al. (2017) and Habte et al. (2022) who found that farmers' perceptions correlated with actual data. Such inconsistency in farmers' perceptions on temperature and actual climatological data could be explained by an observation by Ovuka and Lindqvist (2000) who noted that the timescales of analysis between researchers and farmers is usually different, for example, farmers focus on growing season, resulting in deviations in farmers' perceptions and actual data. With reference to rainfall, the farmers' mostly perceived a decline in rainfall which could not be substantiated by trend data; these results are similar to those of Debela et al. (2015) and Mulenga et al. (2017). It must however be noted that such reduced rainfall perceptions could be linked to increased climate variability and incidences of droughts which have actually been increasing in the region as demonstrated by Jordaan et al. (2019) and Tfwala et al. (2020), and thus climate extremes may be modifying the farmers' perceptions mainly because of difference in actual measurement of rainfall or drought versus the farmers' measurement, where the actual measurement puts emphasis on a meteorological drought whereas the farmers measurement focuses on agronomic drought (Blench, 2006).

Conclusion
The livestock farmers in the Lowveld of Eswatini practise mixed-farming, where they combine raising livestock with growing crops. Livestock farming is dominated by males who are old, and where there are female farmers, the majority are widows. This is because livestock is reared in a household set-up which is dominated by males. Majority of the farmers have at least primary education, and mainly rear cattle, goats and chicken. Furthermore, most of the farmers are knowledgeable about climate change and they obtain most climate-related information through the radio, followed by television, family members and newspaper.
From the empirical evidence, the climate of the Lowveld of Eswatini has significantly changed in the past 31 years. Temperatures have increased, in the southern part in line with what is happening in Southern Africa. In the northern part; however, the temperatures have decreased, and this is opposite to the finding of most other studies. Annual rainfall has high variability but with no significant changes, however, significant changes occurred for a change in rainfall season or monthly rainfall. The farmers' perceive that their climate has significantly changed over the last 20-30 years, and evidence of this change includes: increased temperatures, decreased rainfall, delayed rain season, increased winds, and greater droughts (incidence and intensity). When the empirical temperature and rainfall data was compared to these perceptions, there was incoherent correlation; for example, in the northern Lowveld, temperature trends decreased, and the summer rainfall decreased; whereas in the south temperatures increased. It is likely that the disjointed correlation between empirical trends and the farmers' perceptions were modified by extreme events, especially the recent El Nino . This is because extreme events are known to alter the respondents' perceptions about climate.
The MNL regression indicated that the common predictor variables of the farmers' perceptions towards climate parameters (such as rainfall amount, droughts frequency and rainfall season length) were the farmers' age, the location, and who managed the farm. There was no significant predictor variable for temperature, and this is likely because over 99.3% of the farmers stated that temperatures had risen.
Given that who managed the farm, farmers' age and the location influences farmers' perceptions towards climate change it is recommended that programs to improve farmers' perceptions of climate variability and climate change should be tailored to target farm decision makers, particlualy by their age groups and their area of residence/location inline with the observed area specific climate trends. Moreover the literature provided has proven that farmers' perceptions of climate change influences adaptation; thus, interventions to increase adaptation need to focus on farmers' perceptions of climate, and how it has changed. This can be done by focusing on the common predictor variables which enable better perceptions of the climate and its impacts, such information targeting these variables could be disseminated using the commonly used sources of climate information such as radio, television and newspaper. Moreover, enhancing the farmers' perceptions or climate change knowledge and its associated risks helps the farmers adapt to climate change impacts. The study concludes that some of the perceptions about climate could be influenced by farmers' experiences of extreme climatic events; thus there is, therefore, a need for a refined study on farmers' perceptions of climate and empirical data to include extreme climatic events.