Introduction

Although climate services in the developing world have improved over time, scientific climate forecasts’ (SFs) use has remained low (Nkuba et al. 2021a). SFs are accessible from radio, television, mobile phones (Jost et al. 2015a, b), the internet, agricultural extension agents and farmer-to-farmer networks (Goddard et al. 2010). Development agencies involved in disaster risk reduction and in rural and agriculture development such as World Vision, Care International, Save the Children and Oxfam have also played a role in the dissemination of SFs (ACCRA 2014). SFs are communicated in local languages in Senegal, Malawi and Tanzania (Hampson et al. 2014; Lo and Dieng 2015). Using appropriate local languages to disseminate SFs improves access and use among farmers (Jost et al. 2015a, b).

The main challenge in using SFs by stakeholders such as farmers is their probabilistic nature (Nicholls 1999): what has been predicted may or may not happen. Decision-making based on forecasts with shortcomings raises serious concerns among arable farmers (hereafter referred to as farmers) (Goddard et al. 2001). The limited predictive accuracy negatively influences use of SFs by farmers. Goddard et al. (2001) found that obstacles to using SFs include meteorologists’ failure to provide full information about the predictive accuracy or reliability of the forecasts, and their interpretation as deterministic forecasts. SFs lack the spatial and temporal specificity that farmers are most interested in. The SFs are disseminated in terms of regions or districts which have wide geographical coverage. Thus, predicted weather events may occur in some areas but not in all. Simply producing and disseminating climate forecasts do not make them valuable to farmers. However, Patt et al. (2005) reported that the use of SFs improved crop yields in Zimbabwe. This implies that there is improved farmers’ welfare through utility of SFs.

Farmers’ limited use of SFs has been associated with lack of saliency, credibility, trust and legitimacy (Patt and Gwata 2002; Cash et al. 2003, 2006; McNie 2007). Furthermore, farmers have raised as major shortcomings of SFs poor spatial and temporal resolution with regard to failure to provide forecasts on onset and cessation of rains (Kalanda-Joshua et al. 2011; Nkomwa et al. 2014).

One of the factors that has led to low use of SFs in Africa is low meteorological station density (Medany et al. 2006; UNECA 2011), which has resulted in poor spatial resolution with negative credibility and trust implications. The meteorological station density in Africa is eight times lower than the World Meteorological Organisation’s (WMO) minimum recommended level (Medany et al. 2006); the distribution of the meteorological stations in Africa does not match the existing agro-ecological systems. Meteorological station density is one of the cornerstones of index-based weather insurance (IBWI) (Akter et al. 2016). Increased investment in rural weather station infrastructure not only improves the use of SFs but also increases the uptake of IBWI (Amare et al. 2019) meaning lending institutions may not be in be position to provide much needed insurance and credit facilities to African farmers due to lack of climate information. To address the poor station density in Africa, the United Nations Development Programme (UNDP) has supported a number of African countries by providing them with automatic weather stations (Snow et al. 2016). Besides station density, other challenges of climate services in Africa include dysfunctional stations, or poorly maintained, or low data quality due to low national budget allocation to national meteorological services in many developing countries (UNECA 2011; Snow et al. 2016).

The Uganda National Meteorological Authority (UNMA) provides biannual climate forecasts on the onset and cessation of rains, 10-day rainfall forecasts and rainfall distribution (UNMA 2017b). With support from UNDP, there has been increased investment in weather infrastructure in Uganda (Snow et al. 2016). Alongside the high access to SFs (Jost et al. 2015a, b), the Rwenzori region has a low meteorological station density (Fig. 1), with negative implications for the use of SFs. The meteorological authority collaborates effectively with other agencies involved in weather data generation such as agricultural research stations, wildlife protected areas, national universities and private tea and sugar plantations in rural areas (Snow et al. 2016). Only 5% of rain gauges in Uganda are functional and another 921 rain gauges are needed to attain an optimum level (Isabirye 2017).

Fig. 1
figure 1

Study area map

Full size image

There are multiple sources of indigenous forecasts (IFs) (Roncoli et al. 2002; Kalanda-Joshua et al. 2011; Kolawole et al. 2014; Nkomwa et al. 2014). Sources of IFs include farmer-to-farmer extension network, farmers’ organisation, elderly farmers and farmer’s own observations (Appendix Table 3). There are 10 (abiotic and biotic environmental) indicators that are observed to provide weather and climate information (Nkuba et al., 2020b). These indicators are relevant to farmers in their localities providing forecasts at high spatial resolution ( Roncoli et al., 2002; Orlove et al. 2010; Nkomwa et al. 2014). The temporal dimension of IF provides relevant information on onset and cessation of rains, which is very important in farmers’ adaptation to changes in rainfall seasons (Roncoli et al. 2002; Nkomwa et al. 2014). This suggests that IF has high credibility, trust and legitimacy among farmers. Farmers who use IFs only have expressed interest in using SFs after interaction with meteorologists (Orlove et al. 2010; Mpandeli and Maponya 2013). Climate variability negatively influences the use of IFs (Roncoli et al. 2002; Speranza et al. 2010). In light of the challenges of using IFs exclusively or SFs exclusively, some farmers have resorted to using a combination of SFs and IFs since SFs complements IFs (Ziervogel and Opere 2010; Roncoli et al. 2008; Mpandeli and Maponya 2013). Research has shown that some farmers begin with using IFs and then revise their forecast partially or completely after receiving an SF (Lybbert et al. 2007). There are farmers who revise their IFs after access to SFs and there are farmers who use IFs only and who do not revise their forecasts even after accessing SFs. There is high access to SFs (89%) in the study area. This is attributed to the proliferation of FM radio stations and the use of local languages.

Climate forecasts are used in farmers’ decision-making related to when to plant, harvest and what crops to grow for a given cropping season. The climate forecasts are essential in climate change adaptation (Nkuba et al. 2020a). Farmers use climate forecasts to reduce their vulnerability to the impacts of climate risks such as droughts and floods (Hansen 2002). In bimodal rainfall regions of East Africa, onset determines which cereals are to be grown, with maize being grown during periods of early onsets, while sorghum and millet are grown during periods of late onsets (Mugalavai et al. 2008). Forecasts thus influence the choice of crop enterprises for coming farming seasons. Maize is very sensitive to water stress. For sweet potatoes (Ipomoea batatas), potatoes (Solanum tuberosum) and cassava (Manihot esculenta), water stress leads to poor tuber formation (El-Sharkawy and Cadavid 2002; MacKerron and Jefferies 1986; Yooyongwech et al. 2013), sweet potato weevil infestation (Ebregt et al. 2004, 2007), fruit cracking and blotchy ripening for vegetables (Steduto et al. 2012). The farmers’ use of forecasts for variations in the crop-water requirements greatly influences choice of crop.

Some studies have investigated factors associated with climate forecast use. A study done in Northern Kenya and Southern Ethiopia (Luseno et al. 2003) found that location, education level and access to radio are significant factors in the use of SFs. Another study done in Botswana (Kolawole et al. 2014) reported that age and education level are associated with the use of climate forecasts. The two studies did not include factors such as the language in which SFs are received, livelihood choices, the agro-ecological zone or access to agricultural extension, which the current study takes into account. Access to forecasts seems to have gender dimensions, with men reportedly having more access due to their control of the radio in the rural households (Jost et al. 2015a, b). Jost et al. (2015a, b) found that women in Uganda preferred to receive forecasts from gatherings and community radios, while television was the preferred option for rural women in Bangladesh. A study in south-eastern Kenya (Muema et al. 2018) showed that age, gender, drought frequency, radio and access to improved varieties were factors associated with the use of SFs. A study in eight African countriesFootnote 1 in East and West Africa (Oyekale 2015) showed that education level, climate shocks, radio and gender were factors associated with the use of SFs. These studies did not look at factors associated with farmers’ use of IFs only and both IFs and SFs, hence the need for this study.

The above literature review has highlighted factors associated with use of climate forecasts by farmers. Nevertheless, knowledge gap regarding the effect of crop type, agro-ecological zones and climate risk perceptions on the use of IFs only or both IFs and SFs in farming still exists. The objective of this paper is to investigate factors associated with farmers’ use of IFs only or both IFs and SFs in the Rwenzori region in western Uganda. We answer the question regarding what factors are associated with farmers’ use of IF singly or the combination of both IF and SF. This gap has not been adequately addressed in the international literature.

The consensus globally, regionally and nationally is that climate change is occurring and its negative effects have been acknowledged. Access to climate and weather information therefore becomes imperative to the farming communities which experience the direct vulnerabilities from climate change. In this study, we investigate what can be done to increase use of SFs among rural farmers in Uganda, where there is also a tendency of farmers to use their own non-scientifically based indigenous forecasts (IFs). This study has highlighted to policymakers, climate scientists, climate change activists and government funding partners that farmers who exclusively use SF forecast are very few, with majority of farmers still depending on IFs which are inaccurate and arbitrary. This might have the consequence of farmers not being able to respond to climate change risks appropriately and timely if there is still over reliance on IF forecasts. This study identifies the factors that can be targeted by government and its stakeholders to turn around this situation and increase usage of SF forecasts. This study has also revealed that possibly it would be hard to promote use of SFs only in rural communities but rather a hybrid between SFs and IFs might lead to wider usage of SFs.

Theoretical framework

This study investigated decision-making under uncertainty. Some scholars emphasise the consequentialist approach to decision-making, implying that decision choices have repercussions. The farmers who are the recipients of SFs may or may not use them. National meteorological systems are the principal providers of SFs. The farmers receive SFs which are used to examine decisions under uncertainty. In this study, UNMA disseminates SFs to the farming community. Some of the farmers may not use SFs even after receiving them but use IFs only. There is a budget allocation by the Ugandan government for gathering and dissemination of SFs and early-warning information by funding UNMA. Farmers who use information from meteorological stations are using posterior beliefs which may be objective. Information from meteorology stations can facilitate farmers to change their knowledge (prior beliefs) about climate. Farmers may update their prior beliefs (indigenous knowledge) with information from meteorologists (posterior beliefs) and may follow the recommendations from the experts. Alternatively, the farmers may not believe in the accuracy of the information from the meteorologists, and hence, they may not update their prior beliefs, resulting in their not following the recommendations from the experts. Farmers receiving SFs may change or revise their prior beliefs based on indigenous knowledge, underscoring the importance of confidence in forecast information received (Luseno et al. 2003). The meteorological stations make use of historical rainfall data and other climate parameters to forecast what the rainfall pattern will be in the coming days and months (seasonal climate forecast). Meteorologists go further in recommending what actions the farmers ought to take. However, the final decision on what actions to take lies within the farmers. The value and use of climate forecasts by farmers are based on the range of actions and their capability to respond. This study seeks to fill the knowledge gaps regarding the factors associated with farmers’ decision-making under uncertainty based on the climate forecasts (SFs and IFs) they used. The paper contributes to the current debate on influence of climate information on decision-making under uncertainty.

Materials and methods

The study area

The study was conducted in the Rwenzori region of western Uganda. In terms of climate, the region experiences bimodal rainfall, which contributes to having two cropping seasons with the first season running from February to May and the second season from August to December. The temperature ranges from 12 to 24 °C, annual rainfall ranges from 800 to 3000 mm, and elevation ranges from 500 to 5000 m above sea level.

Most of the farmers have completed primary education (Nkuba et al. 2020a). The farmers’ ability to comprehend climate information can be enhanced through stakeholder engagement by Uganda National Meteorological Authority (Nkuba et al. 2019).

The soils are mainly sandy loam and sand clay loam which support production of cereals, tubers and vegetables (Nkuba et al. 2020a). Tree planting for commercial crop production of coffee, cocoa, fruit trees and commercial woodlots is very vibrant in the study area (Nkuba et al. 2020a). Access to pastures supports climate change adaptation measures for livestock farmers such as livestock migration, herd mobility and livestock diversification (Nkuba et al. 2021c).

The region was selected because it has several agro-ecological zones—mountainous, lowland, mountainous and forested, wetland and forested—in order to investigate the effect of agro-climate on climate forecast use. Wildlife protected areas (WPA) in the study area (Kibale, Toro-Semiliki and Mount Rwenzori National parks) (Fig. 1) and Rwebitaba Zonal Agricultural Research Institute (in Kabarole district) have meteorological stations which provide climate data used in meteorological forecasts for the region. Weather stations are in government-aided establishments such as research stations and WPAs. Farmers in remote areas tend to find SF predictions to be inaccurate. Farming is a major livelihood source in the region. Access to SFs in local languages is due to the spread of FM radio and television stations. The region is inhabited by many ethnic groups, all of which use indigenous knowledge in their day-to-day lives.

Data collection and analysis procedures

Data was gathered from August to October 2015. A respondent survey using questionnaires was used to collect data on socio-economic factors, farm and institutional characteristics by trained research assistants. The respondent survey information was triangulated with data from focus group discussions (FGDs) and interviews. FGDs were used to get farmers’ views on the use of IFs and SFs. FGDs were held in Kyenjojo and Kabarole districts to get the farmers’ views on the use of IF and SF. A female FGD of 15 members and a male FGD of 16 members were carried out in Kyejonjo district. In Kabarole district, a female FGD of 15 members and male FGD of 17 members were conducted. The members were farmers who use IFs and SFs in their farming activities. To ensure that members of FGDs expressed their views freely, gender-segregated focus groups were used. The use of the mixed methods approach was for triangulation of the data obtained to enhance its validity and reliability. Data was analysed using Stata 16 statistical software.

This study is nested in an earlier study (Nkuba et al. 2020a). A multi-stage stratified approach was used in the sampling of respondents. The first stage involved districts, the second stage counties, the third stage sub-counties and the smallest unit the household. The Uganda Bureau of Statistics disseminates the population data of households according to district, county, sub-county and household. The selection criteria were (i) farming systems, namely arable farming, pastoralism and agro-pastoralism and (ii) agro-ecological systems such as forested, lowlands, mountainous and wetlands. The statistically selected sample size of farmers was allocated to particular sub-counties in the selected districts using proportional allocation to size, where the size represents the number of households in the district. Based on a population of the study area of 102,496 households, according to the Uganda population census report of 2014, a sample size of 778 was selected with 95% confidence level and a margin of error of 3.5%. To allow for replacement in the sample of respondents who might drop out of the study, 19% of the statistically selected sample was included, giving a total study sample of 924. This was also to ensure a good sample size for sub-samples (for those who use both IFs and SFs, and IFs only). After data cleaning, 17 questionnaires were excluded due to incomplete responses. Of the 907 respondents, 580 were farmers, 270 pastoralists and 57 agro-pastoralists. This paper limits itself to the 580 farmers in the sample.

Theoretical model

The probit regression was used to analyse farmers’ use of forecasts. The probit model was used for examining the likelihood of a future climate event (for example rain onset and cessation) happening at a particular time using climate forecasts. There is no complete certainty regarding the occurrence of future climate outcomes that have been predicted. The predicted outcome may or may not happen. The predicted climate outcomes (dependent variables) in this study address use or non-use of IFs and SFs. IF and SF are specified as dichotomous outcomes with yes (use) and no (no-use), coded as 1 and 0 respectively. The probit model is a non-linear model that estimates with probabilistic maximum likelihood, often used for binary outcomes (Gujarati 2013). The regression results in this study are based on probit regression estimates. The probit model was estimated using Stata 16 software.

There is potential for presence of self-selection in the use or non-use of forecasts. To overcome issues of selectivity convoluting the estimates, we have controlled for several conditioning factors that could be correlated with use of forecast and at the same time also influence outcomes of interest. We think this approach helps to minimise self-selection bias due to forecast use. We also tried to estimate models with and without forecasts to have a feel for the extent to which the estimates are being affected by selectivity. The estimates did not change when forecasts were excluded in the models. With these two approaches, we think that the possible threat from self-selection is not very serious in our analysis. We could use the instrumental variable approach because it was difficult to find strong excluded instruments that would satisfy all the instrument validity requirements.

Empirical model

The empirical model used in the analysis was specified as follows.

$$\begin{array}{l}{\mathrm Y}_{\mathrm{ij}}=\beta_1\mathrm{Household}\;\mathrm{characteristics}+\beta_2\mathrm{Information}\;\mathrm{sources}+\\\beta_3\mathrm{Farm}\;\mathrm{characteristics}+\beta_4\mathrm{Agro}-\mathrm{ecological}\;\mathrm{zone}+\beta_5\mathrm{Social}\;\mathrm{Capital}+\\\beta_6\mathrm{Percerption}\;\mathrm{of}\;\mathrm{climate}\;\mathrm{risk}+\beta_7\mathrm{Climate}\;\mathrm{risk}+\beta_8\mathrm{Wealth}+\varepsilon\end{array}$$
(1)

For use of IF only

$$\begin{array}{l}{\mathrm Z}_{\mathrm{ij}}=\beta_1\mathrm{Household}\;\mathrm{characteristics}+\beta_2\mathrm{Information}\;\mathrm{sources}+\\\beta_3\mathrm{Farm}\;\mathrm{characteristics}+\beta4\mathrm{Agro}-\mathrm{ecological}\;\mathrm{zone}+\beta_5\mathrm{Social}\;\mathrm{Capital}+\\\beta_6\mathrm{Percerption}\;\mathrm{of}\;\mathrm{climate}\;\mathrm{risk}+\beta_7\mathrm{Climate}\;\mathrm{risk}+\beta_8\mathrm{Wealth}+\varepsilon\end{array}$$
(2)

For use of IF and SF

where Yij(j = 1,2,3,4) representing the four models of using only IF and Zij(j = 1,2,3,4) representing the four models of using both IF and SF (Appendix Table 4).

In this study, we control for several factors that could influence use of information as well as those factors that have been identified as affecting adoption of agricultural technologies in developing countries. Past empirical and theoretical literature has guided our theoretical choice of the factors that we include in our econometric models as explained below. Factors associated with farmers’ use of climate forecasts include education level, age, gender (Roncoli et al. 2002), livelihood choices (Vogel 2000; Ingrama et al. 2002; Patt and Gwata 2002; Ziervogel and Calder 2003; Crane et al. 2010; Haigh et al. 2015; Klemm and McPherson 2017), translation of SFs into local languages (Ingrama et al. 2002; Ziervogel and Downing 2004), source of forecasts for onset and cessation (Roncoli et al. 2002; Haigh et al. 2015), access to credit (Vogel 2000; Ingrama et al. 2002), access to non-farm enterprises (Ingrama et al. 2002; Ziervogel and Calder 2003; Crane et al. 2010; Klemm and McPherson 2017), access to agricultural extension (Vogel 2000; Ziervogel and Downing 2004; Coles and Scott 2009; Crane et al. 2010; Haigh et al. 2015; Klemm and McPherson 2017), access to improved crop varieties (Ingrama et al. 2002), climate risks (Ingrama et al. 2002), agro-ecological zone (Ingrama et al. 2002; Ziervogel 2004; Klemm and McPherson 2017), perception of climate risks (Vogel 2000; Ingrama et al. 2002; Ziervogel 2004; Haigh et al. 2015) and farm size (Vogel 2000; Coles and Scott 2009). Climate risks and perception of climate risks are important factors associated with farmers’ use of forecasts (Vogel 2000; Ingrama et al. 2002; Haigh et al. 2015). Farmers’ cognitive biasesFootnote 2 have an effect on their risk perceptions which influences climate forecasts use (Waldman et al. 2019). Unlike pastoralists, who practice herd mobility as an adaptive mechanism, farmers cannot migrate their crop fields. Changes in the onset and cessation of rains progress slowly towards their final manifestation in damage to crops (Slovic 2000a).

The explanatory variables include Respondent characteristics (H): level of education, age, gender and farming experience; farm characteristics (F): type of crops grown and type of livestock kept; institutional characteristics (I): sources of IF, sources of SF, access to agricultural extension, credit access, improved crop access, non-farm access; agro-ecological system (A): forested, lowland, mountainous, wetland, mountainous and forested; social capital (S): membership of farmers’ organisation, farmer-to-farmer extension networks; perception of climate risks (P) drought increase, flood increase; climate risks (C): flood experience, drought experience; wealth (W): farm size. Variables and expected signs for use of IFs only and both IFs and SFs are shown in Appendix Table 5.

Results

Descriptive overview: use of forecasts

Results show that almost half (49%) of the farmers used IFs only and half (50%) used both SFs and IFs (Appendix Table 6). Only one respondent used SF only. Over half (54%) of the respondents were male (Appendix Table 7). There were no significant differences in the use of either both IF and SF or IFS only between men and women respondents. Forecasts for rain onset and cessation were very crucial in farming. Considering the sub-sample of those who used IFs only, all the respondents used indigenous knowledge in predicting the onset of rains compared to 77% for the sub-sample of those who used both IFs and SFs. Most of the farmers (97%) reported that IF was reliable, compared to 43% for SF. Farmers who used both IFs and SFs reported that they used SFs to confirm the IFs based on what they had observed and learnt from elders. The most important crops grown were cereals and tubers (Appendix Table 3). Farmers’ organisations and farmer-to-farmer networks also played a crucial role in providing climate information (Appendix Table 3). Radio and non-government organisations were important SF dissemination channels (Appendix Table 8). Local FM stations provide SFs in local languages (Appendix Table 7), which has greatly improved access even to farmers with no formal education. Agricultural extension workers played a minimal role in the dissemination of SFs and information on rain season duration. Agricultural extension has been impacted by the involvement of Uganda Peoples Defence Force under Operation Wealth Creation, which plays a major role in the government distribution schemeFootnote 3 for agricultural inputs such as improved seeds, seedlings and livestock during onset of rains. The results show that there were significant differences between farmers who used both IFs and SFs and those who used IFs only (Appendix Tables 3, 7, and 9 at 5% level of significance).

Farmers from mountainous forested areas and bare mountainous areas were using IF only significantly more than both IFs and SFs (Appendix Table 7). Location and terrain appear to affect use of SFs, possibly because of differences in the level of infrastructural development between mountainous areas and lowlands. Farmers in lowlands were more likely to use SFs, while their counterparts in mountains were more likely to use IFs. This suggests constraints linked to quality of infrastructural development limiting access to SFs. Farmers growing cereals were using both IFs and SFs significantly less than IF only (Appendix Table 3). Cereals are very sensitive to water stress and are grown on a commercial scale in the study area. This increases the reliance on rain-fed agriculture in rural areas where there is limited use of irrigation, resulting in an increase in the use of SFs.

Factors influencing farmers’ use of indigenous forecasts only

The results show that factors positively and significantly associated with using IFs only among farmers were as follows: livelihood choices such as tubers and goat production, agro-ecology such as being resident in mountainous areas, reception of information from farmers’ organisations about the onset and cessation of rains access to government programme interventions on climate change adaptations and perceptions of climate variability and change as increases in floods and drought (Table 1). Factors negatively and significantly associated with using IFs only were non-farm access and livelihoods that depend on maize production.

Table 1 Marginal effects of use of IFs only for arable farmers
Full size table

Government programme interventions on climate change adaptations include the provision of improved crop varieties and tree-planting materials to farmers, whose adoption depends on forecasts for the onset of rains. The agricultural inputs, such as seeds and seedlings, were delivered by National Agricultural Advisory Services, Operation Wealth Creation under the Office of the President and the Uganda Forestry Authority. Farmers had confidence in IFs and hence used it for predicting the onset of rains to effectively participate in government adaptation programmes. Cereals (especially maize), tubers and tree crops are sensitive to water stress, especially during the vegetative stage, and farmers’ confidence in IFs influenced their decisions on choice of crop enterprises and allocation of resources, depending on the onset and cessation forecasts. Farmers preferred to grow maize in the second rainy season, which is long enough and runs from August to November. Short-maturing cereals like sorghum and millet were preferred for the first rainy season, which is short and runs from March to May. A key informant reported that winds in the mountainous and forested areas on the windward side of Mount Rwenzori have high moisture content which easily reaches saturation point resulting in rainfall. Farmers in such areas rely a great deal on observing IF indicators like clouds and wind to get reliable forecasts.

Factors influencing farmers’ farmers’ use of both indigenous forecasts and scientific forecasts

The findings show that factors positively and significantly associated with using both IFs and SFs were reception of SFs in local language and English, attainment of higher education (diploma or advanced secondary school education), access to SFs through radio and TV, access to short-maturing crop varieties, availability of agricultural extension services, age and fellow farmer as source of SFs (Table 2).Footnote 4 Factors negatively and significantly associated with using both IFs and SFs were livelihood that depend on vegetables, tubers and maize, being residents in mountainous forested areas, primary school education, Non-Governmental Organisation as source of onset and cessation forecasts and drought experience, and perceiving climate variability and change as seasonal rainfall change and drought increase (Table 2). Vegetable growers seemed not to have enough confidence in SFs as a basis for decision-making, probably because of vegetable crops’ water requirements. Command of English has a higher impact than local language on the use of SFs. Farmers who are comfortable with receiving SFs in English have fewer constraints than those who receive SFs in local languages. English may be more used by elite and well-to-do farmers. Farmers using local languages have less ability to access SFs compared to their counterparts who also use English.

Table 2 Marginal effects of use of both IFs and SFs for arable farmers
Full size table

There are few weather stations in the region (Fig. 1) whose rainfall data is used to make predictions, resulting in poor predictive accuracy and wide spatial variation for forecasts from UNMA. The rainfall data from the weather station in the mountainous area of Kabarole was used to make forecasts for the mountainous and forested area, quite far from the station.Footnote 5 A key informant pointed out that there was apparently no weather station for rainfall data in the mountainous and forested area of Bundibugyo. This casts doubt on the reliability of SFs for seasonal forecasts for mountainous forested areas using extrapolated rainfall data from weather stations located in distant ecological zones. Forecasts were made in terms of regions and districts, yet farmers are interested in information at parish or village level. Receiving SFs in local languages made dissemination accessible to farmers, even to those with no formal schooling. Mountainous and forested areas have high precipitation in the form of mist and fog, which contributes positively to soil moisture availability.

Climate risks like drought are negatively associated with the use SFs. It is probable that climate risks like droughts can influence farmers’ cognitive biases such as availability bias, anchoring effect and confirmation bias, which could impair farmers’ judgement in using SFs. Farmers can overestimate or underestimate the likelihood of droughts due to over-confidence in their recent memory of the severity and frequency of the occurrences. It is also possible to consider long dry spells as droughts.

Climate risk perception such as drought increase and seasonal rainfall change was negatively associated with use of SFs. Climate risk perceptions can influence farmer cognitive biases such as framing effect, hindsight bias, anchoring effect and confirmation bias, which could impair farmers’ judgement in using SFs. The framing effect created by the limited coverage of dissemination mechanisms such as media about the effects of change in rain onset and cessation can lead farmers to underestimate the start and end of the rainfall season. Most media and meteorologists tend to focus more on rainfall distribution in terms of ‘above normal’ or ‘below normal’ in their dissemination of the seasonal climate forecast for the farming season, rather than duration of the rainy season. Farmers’ experiences can make them over-confident and consequently impair their judgement of frequency and severity of droughts.

But there were also respondents who perceive that SFs are not credible, as exemplified by the following observation by a female participant in the FGD: ‘For me, I don’t think it’s truth because there is when they told us to prepare for heavy hail storms but they didn’t come’ (i.e., scientific forecasts are not reliable).

There is also a lack of local specificity in SFs, rendering them less credible to the farmers. One FGD participant, for instance, remarked that ‘You can see that it [rain] has not happened here but when in another place it has’.

Some doubt was also cast on the legitimacy of SFs, as indicated by one participant who declared that ‘They also just guess’. This indicates that farmers do not trust the processes that meteorologists use to make forecasts, and consequently, they distrust the forecasts as well.

Farmers attached high credibility to IFs because of the long tradition of using and depending on them, as is evidenced by the following statement from a FGD participant: ‘from when we were born until we have grown to now it’s what we found our elders using. The elders before us this [IFs] is what they were using, their knowledge of long ago. For that we also grow up following it’.

IFs knowledge is based on the long-term observation of nature and is passed on orally from one generation to another. There were 15 indicators used in IF identified in the study area (Nkuba et al. 2020b) implying that there is a wide variety of sources of forecasts, resulting in increased legitimacy. The indicators are biotic factors such as plants, insects and birds and abiotic factors such as wind and clouds. The abiotic factors are similar to what meteorologists use in their predictions. Farmers’ organisations and elderly farmers were social capital that provide forecast information within their specific local area.

Religion also plays a role in the use of IFs; for example, many respondents said that the onset of rains corresponds to the day of Mother Mary, namely 18 August. One participant in the FGD reported that ‘Indigenous knowledge is a very important thing because that’s the knowledge God created us with. Because even what is taught us [scientific forecasts]… you would learn and by the time you reach home you have forgotten it. But indigenous knowledge is better. Because with scientific forecasts, someone will guess what’s not there’.

Radio is the most widely used method of disseminating information about SFs. Radio stations broadcast in both English and local languages. TV is widely used by elite and well-to-do farmers. TV stations like National TV and Uganda Broadcasting Services are widely used as sources of forecasts by wealthier farmers. A key informant in the civil service stated that agricultural extension workers receive SFs from UNMA through the Ministry of Agriculture on a regular basis.

There are factors that influence the use of both IFs and SFs, and IFs only, differently (Tables 1 and 2). For instance, access to improved crop varieties and agricultural extension services is positively associated with using both IFs and SFs and negatively associated with using IFs only. Access to agricultural extension is closely associated with SFs. Maize production was positively associated with use of both IFs and SFs but negatively associated with IFs for the 5-day forecasts. Short-range SFs such as 5-day forecasts provide reliable climate information (dry spells, or optimum planting and harvesting days) which is relevant to farmers’ involvement in water stress crops like maize.

Being a resident in mountainous and forested areas, and perceiving climate variability and change as drought increase and seasonal rainfall change, was positively associated with using IFs only and negatively associated with using IFs and SFs. Mountainous and forested areas are positively associated with high precipitation due to orographic effects on cloud formation and condensation, contributing to less variability in rainfall.

Having a higher level of education (secondary education and diploma) was positively associated with use of IFs and SFs. Increase in education increases uptake of SFs. However, primary education was negatively associated with use of IFs and SFs. Primary education probably causes the farmers to overestimate or underestimate due to over-confidence based on their recent past experiences.

Reception of SFs from fellow farmers were positively associated with using IFs and SFs, while reception of onset and cessation of rain forecasts from farmers’ organisations was positively associated with using IFs only. Group decisions were more trusted in using IFs only. Confidence in using IFs only is built up through social gatherings of like-minded people. Information disseminated through farmer-to-farmer networks was more trusted in using SFs. This could arise from the confidence that the farmer-to-farmer networks built over time in using SFs with good outcomes.

Discussion

Results showed that significant factors associated with using both IFs and SFs were farm size, education, age, reception of scientific forecasts in local languages, agricultural extension access, short-mature crop access, farmer-to-farmer network and accessing forecasts through mass media. This study shows that IFs were used complementarily with SFs. On the other hand, significant factors associated with using IFs only were livelihood choices such as tuber and goat production, access to government interventions on climate change adaptations, agro-ecological zone and social capital. Research has shown that dissemination of SFs through local languages is the most preferred among rural households (Antwi-Agyei et al. 2021). Dissemination in local languages is usually facilitated by media such as local FM radio stations. The current study has also confirmed that radio and television play a crucial role in the dissemination of SFs. Radio has been reported to be an important dissemination mechanism for SFs in rural regions (Jost et al 2015a, b). This indicates that widening FM radio station broadcasting coverage in Uganda could provide an opportunity for increasing the utilisation of SFs in rural areas. The main shortcoming of using such media as radio and television for dissemination is that there is usually no possibility for feedback from farmers. This is perhaps one of the reasons that farmers seem to have greater confidence in forecasts from their fellow farmers: there is feedback and discussion of experiences and the implications of using such forecasts. National meteorological services need to employ dissemination mechanisms that provide feedback from farmers for improved effectiveness in the use of forecasts. The findings of the current study also show that the onset and cessation of rains forecasts from farmer-to-farmer extension networks had mixed results, with a positive association with use of IFs only and a negative association with use of both IFs and SFs. This shows that farmers had more confidence in IFs than in SFs because IFs have local specificity while SFs have broad spatial and temporal variations (Speranza et al. 2010). Farmers’ indigenous knowledge of climate forecasts is more trusted than forecasts by climate scientists (Kolawole et al. 2014; Roudier et al. 2014). The results of the current study have also shown that farmers have negative associations with use of forecasts from NGOs. The challenge is that NGO staff operating in rural areas may not be able to answer all the farmers’ questions about forecasts, which may engender a lack of trust in the forecasts among farmers (Ofoegbu and New 2021). This study shows that farmers involved in tuber production are more associated with use of IFs only. Farmers involved in tuber and maize production do not have confidence in SFs because of past experiences of poor spatial and temporal specificity and, therefore, in their eyes, of false forecasts (Ziervogel 2004).

Our study shows mixed results regarding the association of education with climate forecast use. Post primary education was positively associated with use of SFs. Luseno et al. (2003) study in Kenya and Ethiopia showed that formal education improved confidence in and access to SFs. Similarly, a study done in Botswana showed that the more educated a farmer is, the more likely he/she will be to use SFs received through the media (Kolawole et al. 2014). However, primary education was negatively associated with the use of SFs. A study done is Zambia reported that one additional year in education reduced farmers’ perception of change in rain onset due to psychological factors (Waldman et al. 2019). This calls for further research on the effect of psychological factors on climate forecast use and a multidisciplinary approach for climate services research.

Climate change perceptions have a mixed effect on the use of forecasts, with a positive association with use of IFs and a negative association with use of both IFs and SFs. Experienced farmers make adjustments in their livelihoods after experiencing a drought, or they make changes in their responses to the onset and cessation of rains because of the likelihood of a recurrence (Slovic et al. 2000b). Farmers experience high crop losses due the change in rainfall duration and severe droughts, thus increasing their confidence in using IFs only. The range of alternatives for farmers is limited, which increases the level of catastrophe to their livelihoods (Slovic et al. 2000c). This also influences the use of IFs only, in order to avoid or to minimise experiencing such risks as loss of seeds due to late onset of rains, or poor harvests due to early cessation of rainfall. Farmers’ assessment of risks, which influences their use of IFs, is related to climate risks that reduce household welfare due to crop yield losses (Slovic 2000, 2000a, 2000b, 2000c). However, the perception of unpredictable rains as climate change is associated with using both. This is consistent with Partey et al. (2018),who found that increased rainfall variability positively influences the use of climate information in Ghana. The seasonal climate and short-range forecasts from UNMA provide predictions of rainfall distribution which influence risk perceptions (UNMA 2017a), as illustrated by a participant in a FGD, who said that ‘Now like this season when they say the rains are going to be a lot, we become cautious of the floods’. This indicates that SFs are also positively associated with risk perceptions and climate change perceptions. The results from the FGDs show that SFs are associated with increased crop yields which is consistent with findings from studies done in Burkina Faso and Zimbabwe (Patt et al. 2005; Roncoli et al. 2008). Research shows that farmers who feel they have adaptive capacity for a particular climate risk (availability heuristic) tend to have lower the climate risk perception (Duinen et al. 2015) which can lead underestimating its effects (Tversky and Kahneman 1979).

Farmers with drought experiences have hindsight bias leading to increased confidence in their degree of prediction of future droughts without using SFs. Drought increase is a dreaded risk whose calamities are known through loss of crop yields and change in rainfall season is a delayed risk with a slow manifestation of harm. The recent memory of climate risks like rainfall seasonal change and drought could influence the use of SFs due to hindsight bias and anchoring effect. The framing effect from various stakeholders and media can result in making rainfall season change look less dangerous than gradual onset of droughts, leading to bias in using SFs. Research has shown that maize farmers have biases related to rain onset (Waldman et al. 2017), which is in agreement with our findings. Research has shown that farmers’ climate risk perception and past experience can lead to poor decision-making related to climate forecast use, due to cognitive biases (Waldman et al. 2019).

Farmers from mountainous forested areas, and mountainous areas, are positively associated with use of IFs only. Due to orographic factors, the wide spatial variations in rainfall in these agro-ecological zones cause farmers to have more confidence in IFs only due to IFs’ local specificity compared to SFs, with SFs’ low spatial coverage because of low meteorological station density, especially in the mountainous areas (Sultan et al. 2020). Furthermore, Leroux (2001) found that ‘with increasing altitude the nature of precipitation changes; raindrops become smaller, showers give way to more continuous rain, and thence to rain, drizzle, mist and fog’ (p. 50). Forests act as windbreakers which disturb surface circulation and have a greenhouse effect that reduces temperature variations (Leroux 2001). Relative humidity in forested areas is high and this consequently leads to forests having a strong influence on rainfall over wide areas (Leroux 2001).

The use of both SFs and IFs can be enhanced by experts engaging in constant dialogue with end-users. Studies in Mozambique, Kenya, Ivory Coast, Senegal, Tanzania and Benin have shown that confidence in SFs greatly improved when national meteorological services received constructive feedback from stakeholders such as farmers (Ziervogel and Opere 2010; Lo and Dieng 2015). Research has also shown that the legitimacy, salience and credibility of SFs have been improved by the active engagement of farmers, resulting in farmers’ increased use of SFs (Bouroncle et al. 2019). Integration of SFs and IFs is critical for the effective utilisation of forecasts in rural livelihoods based on farming. Ziervogel and Opere (2010) have provided information on projects in Africa that had integrated to good effect meteorological and indigenous-based seasonal forecasts in the agricultural sector. Stakeholder engagement can be in form of farm rainfall data generation by farmers. Botswana’s Department of Meteorological services has installed rain-gauges for farmers who send daily rainfall data via telephone. Farmers involved in farm rainfall data generation get excited when rainfall data from their farms is read during the news bulletin on national television station. Botswana meteorologists conducted meteorology extension services during the agricultural shows and held feedback sessions with farmers involved in rainfall data generation before the COVID-19 pandemic. A pilot study on citizen science in Namibia and South Africa has shown that a farmer involved farm rainfall data generation had much trust in SF, constructive engagement with meteorologists, reduced vulnerability to climate change, improved crop yields, livestock sales and household incomes (Landman et al. 2020).

The results of the study are applicable to humid tropical zones and may not be applicable to a semi-arid area. This calls for further research regarding factors associated with use of indigenous and scientific forecasts in semi-arid areas.

Conclusion and recommendations

This study has established that farming in the Rwenzori region is informed by the use of both IFs and SFs or IFs only. Factors that promote the use of SFs include post-primary school education, access to appropriate rural institutions and dissemination of forecasts through radio using local languages. However, primary education was negatively associated with use of SFs. This is attributed to farmers’ cognitive biases. This calls for further research on the effect of psychological factors on climate forecast use. Research on climate information use requires a transdisciplinary approach involving environmental psychologists, social scientists, agronomists and meteorologists.

Farmer-to-farmer extension networks are positively associated with use of IFs only and negatively associated with the use of both IFs and SFs. This calls for active stakeholders’ engagement by national meteorological services. This will improve the salience, legitimacy and credibility of meteorological forecasts in rural areas. Climate risk perceptions and climate risks are negatively associated with the use of SFs. Addressing farmers’ cognitive biases associated with climate risks and climate risk perceptions by national meteorological systems actively engaging stakeholders in climate forecast dissemination mechanisms could improve uptake of SFs.

Investments in ddissemination of weather forecasts should be maintained and or increased, as the study has shown that television and radio have a positive impact.

SFs were found to be complementary to IFs. SFs reinforce IFs. This calls for co-production of climate information in order to promote an increased use of forecasts in rural areas.

Indigenous forecasts will continue to play a major role in influencing land-use interventions among farmers in Uganda and sub-Saharan Africa. We do not contest the use of SFs among farmers because trust in SFs among farmers is likely to improve with an increase in the number and density of weather stations in rural areas. Investment in more weather stations (automatic weather stations and rain gauges) in farming areas is a key factor in obtaining more spatially specific and accurate SFs. This could result in the improved use of SFs which might lead to improved food security and reduce vulnerability to climate change. Governments in developing countries, the private sector and the global and regional development partners should support investments in weather stations and capacity building of national meteorological systems.

These possibilities are of course influenced not only by the use of forecasts but also by a number of other equally important factors such as access to inputs, agricultural extension services and credit.

A longitudinal study on the validity of indigenous forecasts should be explored. In places with bimodal rainfall distribution, data has to be collected for both seasons in a period of 3 years or more to get meaningful outcomes. For comparative purposes, similar research should be conducted in other agro-ecological zones such as semi-arid areas and temperate areas, which were not covered in this study.

The study has also established that farmers use scientific and indigenous forecasts in making decisions under uncertainty. Some farmers receive SFs but do not use them, instead using IFs only. Some receive SFs and update them with IFs, resulting in using both indigenous and scientific forecasts.