**4. Discussion**

In Ethiopia, long-term changes in climate conditions, the inter-annual and seasonal variability of temperature and rainfall, and the frequency of occurrence of extreme events are detrimental for agricultural activities and food security. The results of the analyses of temperature and rainfall time series data reveal a variety of changes in climate conditions of the study areas and notable differences between the agro-ecological settings. The findings generally show increasing warming, annual and seasonal rainfall variability, increasing extreme events, variation in rainfall onset and cessation dates, and convergence and divergence between measured variables and perceptions.

The average temperature of the study areas is increasing which reflects the rising global mean temperature. Such increasing trends of temperature in Ethiopia are also reported in other studies [12,16]. Concerning rainfall, we found a significantly increasing trend in the midland areas but no trend in the highland and lowland areas. Like other parts of Ethiopia [1,12], the study areas are characterized by inter-annual and intra-seasonal rainfall variability. A shift in rainfall anomalies each year indicate the repeated occurrence of rainfall deficits during the farming seasons. In addition, *belg* season is characterized by either total failure of rainfall or false start, both referring to a lack of rainfall to undertake farming activities. Although easterly winds from the Indian Ocean and shifts in the Inter-Tropical Convergence Zone are the main underlying factors for rainfall variation in Ethiopia [10], the diverse topography of the country plays a crucial role in the variability of temperature and rainfall distribution across agro-ecological areas.

The effect of high variability in the amount and distribution of *belg* rain on the livelihood of smallholder farmers in Ethiopia is noticeable for various reasons. First, since the season comes after a long dry season, *belg* rain is crucial for water availability, the production of *belg* crops, and the growth of pasture for livestock. Second, rainfall variability during the *belg* season constrains farmers' options to produce *belg* crops [10]. Although *belg* crops are important for farmers to bridge the time until the harvest of summer crops without significant food shortage, the risk of planting these crops is very high due to prolonged dry spells and short growth periods. As noted by farmers, these result in crop failure, lower crop productivity, and the abandoning of production of *belg* crops, ultimately increasing vulnerability to food insecurity. Third, the poor performance of *belg* rain affects crop production activities during the subsequent main rainy season by influencing the soil moisture and thereby the time of planting long-duration crop varieties such as maize and sorghum [15]. Variability in the amount of rainfall and the time of onset and cessation are also challenging for farmers as they cannot follow conventional farming calendars. Variability or failure of rainfall further exacerbates under- and/or unemployment due to loss of farming days.

There are also challenges associated with reliance on *kiremt* rainfall for crop production. Due to yearly variation in the time of onset and cessation, there is high uncertainty in farmers' decisions of types of crops to be produced and time of planting. In the lowland areas, for instance, owing to the normal or late onset and early cessation, LGP is shorter and the rain stops before the ripening of crops. Consequently, farmers harvest substantially lower yields or there is a complete failure of crops. In the midland areas, too, early termination of rainfall at the beginning of September makes crops infertile. Farmers in the highland areas would benefit from early rainfall and early planting as crops are harvested earlier. However, late onset results in late planting, which makes crops with longer-duration growth periods vulnerable to very cold weather that often starts in September/October and lasts until December, leading to an immense loss of yields.

Consistent with previous findings [2,14,16], the results suggest increasing warm days and nights and decreasing cold days and nights. An increase in extreme events causes changes to human systems much more than changes in average climate conditions [25]. Warming leads to higher rates of evaporation [1] and puts additional stresses on water resources [3], which, through a reduction of crop and livestock production, escalates livelihood vulnerability. There is also a risk of an increase in pests, weeds, and disease which affect both crop and livestock production [5]. The significant values of heavy (R10mm) and very heavy (R20mm) precipitation as well as maximum 1-day (RX1day) and maximum

5-day (RX5day) precipitation denote a high intensity rainfall in the midland areas. The occurrence of flooding, which was mentioned by farmers as one of their problems, is partly explained by the significant increase in heavy precipitation in the area. The effect of flooding is aggravated by the sloping topography of the area and lack of vegetation cover. The occurrence of landslides in the midland areas is also partly related to heavy rainfall. In the lowland and midland areas, frequent occurrences of CDDs and drier years have a deleterious effect on farming activities and farmers' livelihoods.

Both convergence and divergence are observed between farmers' perceptions and the results of meteorological data. Despite heterogeneity among farmers, the perception, of more than half of them, of temperature, the occurrence of drought, and the late onset and early cessation of rainfall was in unison with the meteorological data. There was a clear overlap between the perception that temperature is increasing and the statistically significant increasing trends of temperature data. This finding is congruent with many previous studies that showed consistency between perception and measurement of temperature [45]. However, there was variation, especially regarding rainfall trends. Farmers' perception of decreasing rainfall was not supported by statistical data. We found an increasing trend of rainfall in the midland areas but no significant change in the highland and midland areas. This finding is consistent with previous studies showing that farmers' perception of declining rainfall deviates from rainfall records [19,24,46]. Farmers' perception of trends of rainfall may not corroborate observed meteorological trends for various reasons. As noted in a previous study [20], farmers' perception of decreasing rainfall while it is not happening might show failure in the expected utility and availability heuristic. In line with the utilitarian perspective, farmers' perception of declining rainfall more reflects its livelihood impacts in terms of a decline in agricultural production and food security [18,22,24], which are also caused by factors other than climate change such as a decline in soil fertility and limited use of farm technologies [18,19]. Farmers' perception of declining rainfall might also arise from changes in the seasonality of rainfall and frequency of occurrence of extreme events instead of a change in the total amount of rainfall [46]. For farmers, change in rainfall is perceived as a process, not in terms of quantity [47]. They tend to base their perceptions of recent weather conditions and extreme events as well as on the wrong timing of heavy rainfall instead of long-term changes in average conditions [18,48]. When judging changes in rainfall, the time reference of farmers could be the period when rain is expected for planting, whereas the scientific analysis refers to long term or annual/seasonal changes [49]. Farmers also refer to the amount and distribution of rainfall during the cropping season to form perceptions.

Extreme events such as drought and rainfall variability are more accurately perceived by farmers. Drought takes a central position in the memory of people as it directly affects water and food availability [24], which contributed to a perception aligned with actual measurements. Farmers have good memories of extreme events that perceptions of their occurrence are more likely to be in tandem with observed meteorological data [18]. Although there are farmers whose perceptions deviate from the actual observation, the occurrence of late onset and early cessation of rainfall was correctly perceived by more than half of the farmers. Since the time of onset and cessation of rainfall is strongly related to farming activities, including the preparation of land for planting, farmers are highly likely to correctly recognize these changes. The convergence and divergence between perception and meteorological observation are strongly influenced by the agro-ecological contexts in which farmers undertake their farming activities. This shows that the consistency of perception with observed scientific trends depends on environmental differences in farmers' exposure to different climate variables. Farmers contextually define and characterize the weather conditions of a particular time and place based mainly on what they feel about the cropping season, entailing the important role of perceptual factors in framing their understanding of changes in climate variables.

Household characteristics account for both convergence and divergence between farmers' perceptions and meteorological data. We found that the perceptions of males, older farmers, and those with relatively higher social capital, access to media, and holding a small size of land converge with meteorological data. Male farmers' perception is aligned with meteorological data, which might be

related to their better position to access information and primary responsibility to engage in farming activities. Proper recognition of changes in climate variables is based partly on the number of years of farming experience, meaning that older-age farmers have a more accurate perception than younger farmers [20,50]. Given the complexity of properly observing trends in weather conditions on the one hand and less reliance on traditional weather forecasts in the study areas on the other hand, higher social capital and exposure to mass media facilitate farmers' access to credible information that helps them form a correct perception of changes in local weather conditions [51]. Since their livelihood is most pronouncedly affected by adverse climate conditions, poor farmers are relatively well cognizant of changes in local weather conditions [50]. Conversely, misperceptions were noticed among economically better-off farmers. This is evident from the divergence of perceptions among farmers owning a large size of land and with medium economic status. Economically better-off farmers are more likely to generate their livelihoods from multiple sources that they are less dependent on weather-sensitive livelihood activities. Hence, they are likely to misperceive 'real' changes in climate variables. The results of our study also suggest that a lack of education contributes to the misperception of changes in weather conditions. Lack of education undermines access to varying sources of information and the cognitive ability to process information and make use of it to form an evidence-based perceptions.

Farming decisions and climate risk management plans partly depend on the availability of and access to reliable and relevant weather information. The use of traditional knowledge to forecast weather information is constrained by the high variability of the microclimate that made the forecasts less reliable. In the past, climate change occurred gradually and extreme events occur once in many years so that farmers can develop knowledge systems to adapt to. However, nowadays, the weather condition is highly variable not only between years and seasons but also within a day so that it has become difficult to describe the complex situation using the traditional systems that had been in use in the past. Although this is partly addressed through access to media which help farmers to have an accurate perception of changing weather conditions, there are also limitations associated with access to modern weather information. Weather stations are limited in number and unevenly distributed [12], the result of which fails to clearly show spatial differences of the micro-climate. Since the forecast is also made at a higher spatial scale and on a seasonal basis [52], it is less useful for farming decisions at the local level due to highly diverse topography. In addition, there is a lack of information on the time of onset and cessation of rainfall, which is important for decisions on planting time. Farming and adaptation decisions in an uncertain environment and without access to specific and reliable weather information are challenges for risk management. Besides, the lack of specific meteorological information contributes to farmers' incorrect perceptions of local weather changes [24].
