Application of Agent-Based Modeling in Agricultural Productivity in Rural Area of Bahir Dar, Ethiopia

Abstract

Effective weather forecast information helps smallholder farmers improve their adaptation to climate uncertainties and crop productivity. The main objective of this study was to assess the impact of weather forecast adoption on crop productivity. We coupled agent-based and crop productivity models to study the impact of farmers’ management decisions on maize productivity under different rainfall scenarios in Ethiopia. A household survey was conducted with 100 households from 5 villages and was used to validate the crop model. The agent-based model (ABM) analyzed the farmers’ behaviors in crop management under different dry, wet, and normal rainfall conditions. ABM results and crop data from the survey were then used as input data sources for the crop model. Our results show that farming decisions based on weather forecast information improved yield productivity from 17% to 30% under dry and wet seasons, respectively. The impact of adoption rates due to farmers’ intervillage interactions, connections, radio, agriculture extension services, and forecast accuracy brought additional crop yields into the Kebele compared to non-forecast users. Our findings help local policy makers to understand the impact of the forecast information. Results of this study can be used to develop agricultural programs where rainfed agriculture is common.

1. Introduction

Improvement of crop productivity requires effective use of resources such as water, land, agriculture inputs, and weather information. Climate change coupled with water and food insecurity makes agriculture more challenging. Even though there have been technological advances in crop and water management, weather and climate are still important factors in agricultural productivity [1].

1.1. Maize Production in Ethiopia

Ethiopia is the fourth-largest maize producer in Africa and maize is a dominant production in almost all regions of Ethiopia [2]. There are more than 2.67 million smallholders in the Amhara region [2], which contributes 28.6% of maize production in the country [3]. Besides a small marketed surplus, 75% of all maize produced is consumed by farming households [4]. A total of 16.7% of the national calorie intake is from maize [5], and the average maize consumption is 45–50.5 kg per person/year [6,7]. Thus, maize is an important crop for food security and the main economic activity in the Amhara region [8]. Maize will continue to be a fundamental cereal of households, particularly in rural areas [9].

1.2. Maize Productivity in Ethiopia

Maize has a higher yield potential per hectare compared to other cereals. Average yield for maize is 4.5 tons/ha in the world, whereas there is an average of 6.2 tons/ha in developed countries and 2.5 tons/ha in developing countries. National average productivity in Ethiopia is about 3.0 tons/ha [10]. There is an estimation that current maize productivity could be increased twofold with higher quality and more efficient use of agricultural inputs such as fertilizers, chemicals, and seeds [2]. However, despite government development programs to increase productivity, low productivity still persists as a major agricultural challenge in Ethiopia [11]. Geta [12] states that technological advances through research have not increased efficiency and yield productivity.

1.3. Agricultural Inputs

Seeds are an important input into an agriculture farming system, and the number of seedlings used influences the productivity [13,14]. Farmers use urea as a source of nitrogen (N) and diammonium phosphate (DAP) as a source of phosphorus (P) or hybrid varieties of fertilizers. Recommended rates of fertilizer applications range from 100 kg/ha to 200 kg/ha but vary based on location due to soil fertility, fertilizer types, and altitudes of humid environments [4]. Studies show the application of herbicide positively increases maize productivity [15,16,17]. Soil tillage is also the foundation of crop production systems and efficient impacts on maize productivity.

Agricultural drought is a persistent problem in the Amhara region, where eastern parts of the region experience repeated droughts [18]. There have been four droughts with varying levels of severity between 1975–2003 in the Amhara region [19]. A total of 48 out of 105 woredas (districts in Ethiopia) are drought prone and food insecure. These food-insecure woredas meet only 62% of the daily 2100 kcal per capita food requirements [20]. Irrigated agriculture is negligible, and the region is heavily dependent on rainfed agriculture. Thus, drought is often correlated with food shortages and famines [21].

1.4. Need of Seasonal Forecasts in Agriculture

Climate change increases drought and flood occurrences because of temperature rise and uncertain rainfall conditions. Smallholder rainfed farmers are more vulnerable to such changes [22,23]. Successful seasonal forecasts are considered to bring social and economic benefits to the smallholder farmers in Ethiopia [22]. However, building farmers’ trust in weather forecast information is challenging because of farmers’ susceptibility to incorrect forecasts [24,25]. Poor forecast systems can be damaging and may result in harvest loss. Studies suggest that a weather forecast system should be correct at least 60–70% of the time to increase farmers’ net agricultural income and provide better adoption rates [26,27,28,29,30,31].

1.5. Adoption of Seasonal Forecasts

Weather information has been shown to improve resilience of agriculture and farmers’ abilities to adapt to climate change [10,32]. Seasonal forecasts help farmers improve nutrient management such as fertilizers, chemicals, and other input timings, as well as planting schedules such as sowing period and crop selection [33,34,35,36,37,38]. Thus, farmers should change their decisions with respect to farming strategies to gain benefits from seasonal forecast information [39]. For example, farmers considered receiving rainfall predictions and drought warnings an important factor in farm decision making in Zimbabwe [40]. For better adoption, farmers should be supplied with contextual and location-specific weather forecasts at the correct time [41]. Economic assessment of adoption of weather forecast services revealed an increase in crop yields in India. Mabe [42] stated that having access to weather forecast information was an important step to improving crop productivity and farmers’ livelihoods in Ghana and Senegal.

1.6. Agent-Based Model in Agriculture Management

Although weather forecast information is expected to improve the agricultural sector of Africa and the capacity of farmers to manage climate-related risks, the use of seasonal climate forecasts still remains low [43,44]. Other main reasons for limited adoption of weather forecasts include lack of awareness of the potential benefits and lack of trust in the information source [39]. Thus, assessment of weather forecast impacts is challenging because they are not commonly used in developing countries [45]. Although crop models have been used to evaluate the efficiency of weather forecasts, assessing weather forecasts requires integration of social and environmental systems [46,47]. Agent-based modeling (ABM) is a new approach to studying human–environment relationships to simulate weather forecast uptake by farmers to improve agricultural productivity [48]. Recent computational advances have enabled the application of ABM in several studies of innovation diffusion in agricultural systems [49]. Researchers [50] assessed agricultural policies through ABM to study farmers’ interactions for structural changes such as farm expansion or reduction. Other researchers applied ABM for agricultural policies [51,52,53,54] and a spatial decision support system [55]. In societies, information flows through different channels such as radio broadcasts and government extension workers as well as between family, friends, schools, and other social networks. The diversity of climate information pathways makes it complex and challenging to track sources and information dissemination in communities [56]. ABM provides the ability to couple social and environmental models to study such complex societies and the emergence of collective responses to different management strategies. A distinctive feature of ABM is to model individual agents’ decision-making behaviors while interacting with agents in the community [57]. Thus, ABM performance in farmers’ interactions and farm structural change are found to represent accurate human behavior [58,59]. ABM can also be used to couple farmer decisions with crop models to explore how farmers uptake information and adapt to new agricultural policies.

1.7. Agriculture Crop Models

Agriculture productivity simulation models of soil, crop, and atmospheric systems help in the understanding of the processes of identifying how crops respond to different farmer decisions throughout the cultivation period. The Decision Support System for Agrotechnology Transfer (DSSAT) is a crop simulation model that can simulate crop growth, developmental stages, and yield as a function of the soil, plant, and climate of 42 crops [60]. DSSAT models require daily weather data, soil and soil profile data, cultivar characteristics, and crop management data as inputs. Crop development simulations can be conducted at daily or hourly time steps. At the end of the iterations, each input variable, such as crop, soil, water, and fertilizers, are updated along with crop phenological stages. Another option DSSAT offers is to conduct “what if” scenario experiments for crop management to allow users to vary crop, weather, and management variables [60]. This feature of DSSAT allows the model to be coupled with an agent-based model. The DSSAT is known to be an advanced crop model [61], and DSSAT-CERES-Maize is a dynamic simulation model that helps to understand the maize physiological process that is widely used in the world [62,63]. Accurate crop yield forecasting is an essential goal of many researchers and decision makers throughout the world.

The main objective of this study was to assess the impact of weather forecast information adoption and productivity improvement. Our approach was to (1) identify the main crop grown in the Kebele, (2) study crop productivity of farmers under different rainfall conditions, (3) assess the impact of agents’ forecast-adoption rates into Kebele crop productivity and (4) explore ways to optimize forecast adoption. In this study, ABM was used to determine the adoption rates of farmers. Those adoption rates were then used in the agriculture simulation model to understand crop responses to different farmer decisions. We consider farmers’ planting data, fertilizer use, herbicide use, and tillage frequencies under more and less rainfall conditions.

2. Materials and Methods

2.1. Description of the Area

Rim Kebele is located in Mecha Woreda, West Gojjam Zone of Amhara Region, Ethiopia [64] (Figure 1). We chose this Kebele due to the high reliance of farmers on using rainfed agriculture systems.

There are 18 villages in the Kebele (

Table A1. Information about Rim Kebele villages, household sizes, and cultivated area.

Name of the Village		Estimated Household Size	Cultivated Land Out of Total Area (%)	Forest Land Out of Total Area (%)	Irrigated Land Out of Cultivated Land (%)	Irrigation User Households Out of Total Households (%)
1	Angut Adis Alem	45	81.5	4.2	0	0
2	Angut–Mahal	65	76.8	9.2	0	0
3	Shafri	65	81	4.2	0	0
4	Wuren-1	65	81.8	4.2	0	0
5	Wuren-2	60	81.8	42	0.05	10
6	Wuren-3	60	81.8	4.2	0	0
7	Cheba-1	60	81	4	0.05	10
8	Cheba-2	55	81	4	0	0
9	Deber Mender-1	55	81	4	0.05	10
10	Deber Mender-2	48	81	4	0	0
11	Babo Bate-1	45	76.5	9	0	0
12	Babo Bate-2	45	77	9	0	0
13	Kuyu	50	81	4.2	0	0
14	Ko Rim	54	82	4	0	0
15	Lay Gult	50	82	4	0	0
16	Sendi	54	80	5	1	15
17	Dima	60	81	5	1	15
18	Ketema	387	2	10	0	0
	Rim Kebele	1323	1916 ha (81.8%)	100 ha (4.2%)	20 ha (0.1%)	22%

) and the Partnership for International Research and Education (PIRE) project research team randomly selected 5 villages for this study: Angut-Mahal, Cheba-1, KoRim, Sendi, and Dima. Demographic characteristics of households from the field survey indicated that the age of household head varies between 20 and 81 years old. Education level of the smallholder farmers shows only 28% of farmers can read and write and completed grades 1–12 of the school system, whereas 77% of the farmers are unable to read and write. Average family size of the farmers is five people per household [64]. The climate of Ethiopia varies geographically [65] and is classified into five climatic zones based on elevation and temperature [66]. The geographic location of Rim Kebele is at 2000 m elevation [67] and in a “Weina Dega” climatic zone with an average of 800–1200 mm rainfall depth [68]. Four main seasons known as Kirempt, Belg, Bega, and Tsedey occur throughout the year. Kirempt is the summer season and is the main rainy season for farmers to cultivate their crops from June to September [69]. The average amount of land of the smallholder farmers is 0.75 ha. Farmers tend to prefer traditional weather forecast methods such as looking at clouds (55% of farmers) and wind directions (58% of farmers) compared to scientific weather forecasting methods broadcasted on TV and radios and in newspapers [64].

2.2. Sampling and Data Collection

PIRE researchers visited Rim Kebele in 2017 and summer 2018 to design survey questions and complete preliminary testing. Village-level experts and some household representatives were invited to discuss household data collection methods. A total of 100 households were randomly selected from 5 villages. Survey questions were reviewed in detail with local experts and farmer representatives. The household survey was used to determine demographic characteristics, use of weather forecast information in daily farming activities, crop productivities, and other socio-economic characteristics.

2.3. Crop Pattern

Data from the field survey gave the crop patterns of 100 farmers in Rim Kebele. Survey analysis indicated that maize is the foremost cultivated crop under any rainfall conditions, followed by finger millet and wheat. Barley and teff were found to be less cultivated (Figure 2). Thus, we selected maize for crop modeling and to assess its productivity.

2.4. Farmers’ Decisions on Farming Activities under Different Rainfall Conditions

Based on survey analysis, rainfall predictions lead to changes in farmers’ behaviors. We found 75% of farmers plant maize later if a shortage of rainfall conditions is forecasted. Conversely, 71% of farmers plant maize earlier if a surplus of rainfall conditions is forecasted. Farmers do not change seed types under shortage and surplus rainfall conditions. Farmers reported that seed price also does not change for both conditions. Amounts of 27% and 35% of farmers reported using more fertilizers when there are fewer and more rains, respectively, but the majority of farmers, 62%, do not change the quantity of fertilizers no matter what kind of rainfall they face in a crop season. More farmers, 61%, use fewer pesticides if more rainfall is expected, and 45% use more pesticides if less rainfall is predicted. Amounts of 22% and 24% of farmers reported tilling their land more times if they expect less and more rainfall, respectively (

Table 1. Farmers’ activities under shortage and surplus of rainfall.

	During Shortage of Rain, Farmer:			During Excess Rain, Farmer:
	No change	Earlier	Later	No change	Earlier	Later
Planted maize early or later	12	13	75	19	71	10
	No	Yes		No	Yes
Changed seed type	48	52		56	44
	No change	Less/ decreased	More/ increased	No change	Less/ decreased	More/ increased
Seed price	71	6	23	72	7	21
Used more/less fertilizers	63	10	27	62	3	35
Pest problems	43	12	45	33	61	6
Used more/less pesticides	44	9	47	32	52	16
Tillage frequency	63	15	22	54	22	24
Cultivated less/more land	91	5	4	88	6	6
Grew fewer/more cash crops	80	20		75	5	20
Price of maize	64	4	32	70	21	9
Stored more/less grains	55	30	15	59	16	25
Procured less/more livestock	95	5	0	93	1	6

).

Our study shows that increased tillage does not increase crop productivity (Table 1). Size of cultivated land area does not change in both rainfall conditions. Some farmers, 32%, report that price of harvested maize increases under the less rainfall condition. This increase in price may show a relationship between decreased productivity and decreased rainfall. A total of 21% of farmers believe the price of maize decreases if more rain is expected. However, a majority of farmers (64% and 70%) think the price does not change in both conditions. A total of 25% of farmers store more maize during the more rainfall time, and 30% of farmers reported that they store less yield during the less rainfall time. Farmer’s procurement of less or more livestock does not change in both conditions (Table 1).

2.5. Coupled ABM and Crop Models

Farmers were surveyed about crop patterns (Figure 2) and farmers’ decisions during less and more rainfall conditions (Table 1). Questions such as “what do you do when there is less rainfall?” and “what do you do when there is more rainfall?” were used, and the interviewer listed the farming activities and farmers’ behaviors as responses to such rainfall conditions. We used agents’ adoption rates [64] coupled with a crop productivity model to assess the impact of farmers’ decisions on agricultural productivity (Figure 3). Farmers’ decisions such as planting date, seed type, fertilizer use, pesticide use, and tillage frequencies were used to crop model (DSSAT). Crop modeling programs helped to conduct different treatment scenarios varying farmer decisions to study maize productivity.

2.6. ABM Model

Survey analysis showed that there are actually four types of agents who are the leading communicators in the dissemination and uptake of weather forecast information in the communities (Figure 4). An agent-based model was developed based on the household survey to assess the impact of seasonal weather forecast information used by smallholder farmers. ABM helped calculate adoption rates based on the information flow through these main agents and reported trust values from [64] (

Table 2. Information flow through agent types.

Information Flow		Neighboring Farmer	Other Farmers	Extension Worker	Media (Radio)	Average
Source of information farmers receive from:	Yes	41	36	50	34	40.2
	No	59	64	50	66	59.8
Level of trust in weather forecast from:	Not at all	59	64	50	67	60.0
	Somewhat	34	32	24	23	28.2
	Very much	7	4	15	6	8.0
	Fully	0	0	11	4	3.7
How often agents act-upon the forecast information from:	Never	61	64	52	67	61.0
	Twice a year	32	33	40	24	32.2
	Every month	7	2	7	5	5.2
	Once a week	0	1	1	4	1.5

).

Influential weather communicators are neighboring farmers, other farmers in the Kebele, agriculture extension workers, and radio. Farmers interact with one another, may trust each other, and act upon the weather information. Agriculture extension workers have higher influence levels in the Kebele compared to other agents (Table 2). Based on the survey reports, there are four extension agent workers in Rim Kebele. They are the Kebele agricultural office head, Kebele development agent for natural resource management, Kebele development agent for crop production and management, and Kebele development agent for animal production and breeding [64].

Researchers developed an ABM model to study the dynamics of information exchange among the agents. They coined “Likelihood to Adopt”, or LTA, which is a basic unit of measurement in the range of 0 to 100 LTA points for the model development that each individual agent gains, holds, or loses during interactions with other agents. It is a measure of an agent’s likelihood to receive, trust, and then adopt the weather forecast information [64]. Researchers took the average forecast accuracy, 65%, based on the studies considered, trust, and adoption levels in the community. The 65% forecast accuracy falls within the 55–70% range required for a wide scale adoption of weather forecast information [28,29,31].

In the model, agents interact with their neighboring farmers within a predetermined radius. Studies show diverse learning processes through observing the performance of neighbor farmers, friends, and relatives [70]. Agents also communicate with other agents through networks because researchers found more effective weather forecast information dissemination [71]. Farmers are connected to other farmers with network connections, and this enables modelers to assess how strongly farmers become linked to other distant farmers. Another agent is an agriculture extension worker who has positive impacts in interpreting and conveying the information to farmers [72]. In the model, extension workers have a range of influence levels and directly communicate with farmers to spread the information. Influence level is the measurement of the agent effectively delivering the information to other farmers in the community. The last important communication agent is radio, which delivers the weather information to farmers [73].

Researchers calibrated the model with calibration parameters from calculating the probability levels of information source, trust in that source, and adoption levels from the survey data. Each agent’s adoption likelihood is weighted according to the reported trust levels [64]. Model sensitivity analysis of trust values was conducted through the trust value stratification method used in [74]. Adoption rates were chosen based on sensitivity analysis over different likelihood to adopt points. Later, experiments were conducted varying variables of agents’ intercommunication with other agents such as farmers, agriculture extension workers, other farmers in the Kebele, and radio. Experiments revealed the agents’ potential adoption rates of weather forecast information. However, Musayev et al. [64] did not study the agricultural productivity of these adoption rates. In this study, farmers’ behaviors surrounding farming activities and decisions related to maize yield productivity under different rainfall conditions is evaluated.

2.7. Crop Model

We used Climagen SC to generate weather data such as solar radiation, wind, minimum temperature, maximum temperature, and rainfall for the Bahirdar area based on historical weather data (Figure 5). We retrieved available daily historical weather data for 1961–1988 from NOAA Global Historical Climatology Network V2 [75]. Normal rainfall occurrence is about 1300 mm of rain per year. The more rainfall condition is about 1900 mm and the less rainfall condition is about 900 mm of annual rainfall observed [69].

Rainfall patterns are directly converted into the weather station section of the DSSAT program to set up field conditions. Soil data are retrieved from the world soil database [76].

2.8. Crop Model Calibration

We employed the DSSAT’s CERES-Maize crop model to calibrate and analyze crop productivity under different farmers’ decisions on planting time, usage of fertilizers, herbicides, and tillage frequencies. Based on the rainfall conditions, we set up the planting dates as: early on May 1, normal on June 1 and late on June 25. We used survey data to calibrate the crop model. The average yield is 3462 kg/ha in the survey data. We also analyzed the fertilizer use, herbicide as a chemical use, and tillage number (collectively FCT) of farmers from the survey for calibration of the crop model (

Table 3. Farmers’ decisions in use of farming activities and inputs.

Farming Activities	Used Amount
Seeds used, kg/ha	4–20	20–30	30–50
Crop productivity, kg/ha	2662	4286	5121
Fertilizer NPS used, kg/ha	65–100	100–200	200–270
Crop productivity, kg/ha	3034	4077	3980
Herbicides used, L/ha	0.5–0.9	1–2	2–3.2
Crop productivity, kg/ha	3263	3346	4744
Tillage frequencies	1–4	5–9
Crop productivity, kg/ha	4295	4013
Average crop productivity, kg/ha	3462

).

2.8.1. Seeds Amount Used

We referred to survey productivity data for a normal rain year to calibrate the crop model. Some variables of farmers’ behaviors include the amounts of seeds, fertilizers, chemical inputs, and tillage frequency to produce maize in their farmlands. Less than 20 kg/ha of seeds amount used produces an average of 2662 kg/ha yield, but 30–48 kg/ha of seeds amount used produces an average of 5121 kg/ha. One can see the more seeds sowing increases crop productivity. We used 24 kg/ha of seeds for our crop model calibration (Table 3).

2.8.2. Herbicide Use

Based on the survey, farmers’ usage of herbicides shows a positive increase in yield. For crop model calibration, we used 1 L/ha herbicide use for normal conditions and increased and decreased the usage for different rainfall conditions (Table 3).

2.8.3. Fertilizer Use

Fertilizer use is important in crop productivity. Survey analysis shows a slight increase, 34.4%, in crop productivity under application of fertilizer (NPK) between 65–100 kg/ha and 100–200 kg/ha. However, the average trend of productivity increase does not progress when farmers used an average of 200–270 kg/ha of fertilizer. Thus, for calibration we used fertilizer applications within the range of 100–200 kg/ha (Table 3).

2.8.4. Tillage Frequency

Based on the survey analysis, maize productivity does not increase after 5 times of soil tillage. Thus, we used 1–4 times for model calibration.

Average yield productivity aggregating all farming activities is 3462 kg/ha in Rim Kebele (Table 3). Thus, we calibrated the crop model to reach this value considering seeds amount used, fertilizer use, herbicide use, and tillage frequencies.

2.9. Crop Model Accuracy and Prediction

The CERES-Maize model is a reliable maize model and has been able to accurately predict yield variability and nutrient uptake [77,78]. Researchers around the world examined its level of accuracy and ability to forecast crop growth, grain yield, and nutrient demand in variety of environments. Evaluations require comparing the simulated model outputs with real data for specific yield predictions. Soler et al. [79] applied CERES-Maize for planting-date evaluation and yield forecasting in several regions of Brazil. Their results showed that the model was able to simulate crop phenology and yield accurately with normalized RMSE less than 15%. They stated that accurate yield forecasts could be given at about 45 days prior to harvest date, which is promising for farmers and decision makers.

Gungula et al. [80] applied CERES-Maize for predictions of maize phenology under nitrogen-stressed conditions in Nigeria. Their results show that the model can be reliable for accurately predicting maize phenology, especially days of silking and maturity, grain-filling period, and nitrogen uptake. Mubeen et al. [81] studied model performance in semiarid Pakistan. They found general satisfactory agreement between observed and simulated values for two hybrid seed types (R² = 83% for Monsanto-919 and R² = 84% for Pioneer-30Y87). The crop model showed variability of 19.6–19.9% and a strong positive and linear relationship for grain yield under various treatments and was useful in providing information for decision makers at the farm level in a semiarid environment. Chisanga et al. [82] evaluated CERES-Maize for planting dates and nitrogen fertilizer in Zambia during 2013/2014 cropping season. Planting date significantly affected grain yield at p < 0.05, and the coefficients of variation for the grain yield was 9%. The model’s simulation grain yield was fair: NRMSE = 21.4%. The researchers found that their results showed that the model can be used to accurately determine optimum planting date, yield, and nitrogen fertilizers with reasonable accuracy. MacCarthy et al. [83] used CERES-Maize for farm management practices as a decision tool in the Northern Region of Ghana. The simulated yields were positively correlated with the observed data: R² = 0.73 and Willmott’s d-index = 0.68. Thus, for the last 30 years DSSAT has been used to simulate crop management options to assess associated risks. This model continues to be the most used in many countries and remains the fundamental maize model [84].

3. Results and Discussion

Farmers’ interactions on weather forecast communication has an impact on crop productivities. We applied adoption rates to community-level production. There are 1323 households in the Rim Kebele, and an average farmer’s plot size is 1.45 ha. It can be seen that 80% of the farmers plant maize as their primary crop irrespective of rainfall years (Figure 2).

For normal rainfall and farming activities such as fertilizer use, chemical inputs, and tillage frequencies (FCT), the normal crop model is calibrated to 3466 kg/ha. We took this as a baseline productivity to see the impact of yield changes under different treatments such as planting early, normal, and late under more rainfall or less rainfall conditions (Figure 6).

Crop model results indicated that farmers who plant crops with average normal farming activities under more rainfall conditions produce 3598 kg/ha, which is an increase of 132 kg/ha. From the figure, farmers planting early brings an additional 1091 kg/ha of crop under more rainfall and 601 kg/ha under less rainfall conditions. This implies that farming decisions based on weather forecast information improve yield productivity from 17% to 30% under dry and wet rainfall years, respectively.

Farmers experience a decrease of 105 kg/ha if they plant late under more rainfall conditions and a decrease of 396 kg/ha (

Table 4. Farmers’ decisions of planting time and usage of fertilizer, herbicides, and tillage frequencies under different rainfall conditions.

	More Rainfall Condition			Less Rainfall Condition
Farmers’ decision	Plant normal	Plant early	Plant late	Plant normal	Plant early	Plant late
FCT normal	3611	4296	3285	3072	3096	3900
FCT more	3883	5574	3607	3432	3269	4500
FCT less	3300	3800	3190	3072	2845	3800
Average	3598	4557	3361	3192	3070	4067
Diff. from baseline	132	1091	−105	−274	−396	601

) under less rainfall conditions. Figure 6 shows only positive productivity if farmers plant late if the year is expected to be dry or plant early if the weather forecast predicts a wet year.

3.1. Farmers’ Interactions with Nearby Neighbor Farmers

Farmers’ community-added yield, measured as maize in kg, increases proportionally as interaction radius distance increases. It is recommended that farmers interact not only with his/her direct neighbors (15 unit or 700 m radius distance) but engage intervillage interactions in at least 1200 m distance (25 units) [64]. This would add about 300,000 kg (Table 2A) of more maize into the Kebele production if farmers plant early under more rainfall conditions or about 170,000 kg (Table 3A) if they plant late under less rainfall conditions (Figure 7).

Gurmu et al. [7] stated that per capita consumption of maize is 50.5 kg/ha in Ethiopia. Thus, this would feed about 3500 to 6000 more people if farmers use weather forecast information and respond properly under more or less rainfall conditions. Every unit of distance, or about 50 m, would add 5000 kg to 8600 kg of maize or feed 100–170 more people in the Kebele (Figure 7). Some studies show that agents grasp more pieces of information from neighbors [85,86,87,88]. Farmers’ interactions through workshops similar to our study showed that farmers’ yield increased by 9.4% compared to non-forecast users in Zimbabwe [89]. Farmers preferred to receive seasonal climate information through engaging with other local farmers in Benin, West Africa. Being a member in a farmers organization is likely to increase benefits from weather forecast information by about 14.7% due to adoption, training, and education on best farming practices [39], which is another example of the benefit of farmers’ interactions with nearby farmers in the community.

3.2. Farmers’ Interactions with Other Farmers in Their Network

Farmers interacting through their network connections results in 23% more adoption rates if they are at least twice more connected [64]. Doubling the farmers’ interactions would also bring about 175,000 to 300,000 more kg of maize under favorable conditions if farmers behave according to weather forecast information (Figure 8).

Each incremental network connection improves farmers’ connectivity and brings about 9000–15,000 more kg of maize into the community, or feeds an additional 180–300 people. In other words, every time a farmer adds another farmer into his/her network, he/she supplies additional food for 180–300 people (

Table A4. Adoption rate of farmers’ interactions with other farmers through networks, crop productivities of adopted numbers, and number of people able to be fed under less rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed by the Produce Added
# of Link	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
1	24	2.88	3.12	18	−1144	−1792	15,677	−8352	−13,077	114,434	−165	−259	2266
5	30	3.6	3.9	23	−1430	−2239	19,597	−10,440	−16,346	143,043	−207	−324	2833
10	35	4.2	4.55	26	−1669	−2613	22,863	−12,180	−19,070	166,883	−241	−378	3305
15	37	4.44	4.81	28	−1764	−2762	24,169	−12,876	−20,160	176,419	−255	−399	3493
20	37	4.44	4.81	28	−1764	−2762	24,169	−12,876	−20,160	176,419	−255	−399	3493
25	37	4.44	4.81	28	−1764	−2762	24,169	−12,876	−20,160	176,419	−255	−399	3493

and

Table A5. Adoption rate of farmers’ interactions with other farmers through networks, crop productivities of adopted numbers, and number of people able to be fed under more rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed by the Produce Added
# of Link	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
1	24	5	17	2	873	26,948	−367	6371	196,703	−2676	126	3895	−53
5	30	6	21	3	1091	33,685	−458	7963	245,879	−3345	158	4869	−66
10	35	7	25	4	1273	39,299	−535	9291	286,859	−3902	184	5680	−77
15	37	7	26	4	1346	41,545	−565	9822	303,251	−4125	194	6005	−82
20	37	7	26	4	1346	41,545	−565	9822	303,251	−4125	194	6005	−82
25	37	7	26	4	1346	41,545	−565	9822	303,251	−4125	194	6005	−82

). Advantages of diffusion of new information through social networks are studied by many researchers [90]. Social network messages allows farmers to communicate with peers instantaneously with queries and has the potential to allow them to build trust with one another. Singh Nain et al. [91] studied farmers obtaining immediate advice on crop damage, pesticide application, increasing yield, and crop rotation through WhatsApp messages among 130 farmers from January 2016–December 2017 in Punjab, India. They found that farmers shared on average 13 messages and videos per week, out of which 75.6% were related to agricultural purposes. Social networks, being potential sources of socialization helping the circulation of information, can make farmers aware of the benefits of weather forecast information.

3.3. Influence of Radio in the Kebele

Farmers prefer radio to any other media to receive the weather information in the Kebele. We found improving the radio influence by 60% is sufficient to achieve high adoption rates of weather forecast information in the community [64].

An improvement of 60% in radio influence level would bring about 180,000 to 320,000 kg of more grain or equivalent food to 3500 to 6500 people (

Table A6. Adoption rate of influence of radio, crop productivities of adopted numbers, and number of people able to be fed under less rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed by the Produce Added
Radinf	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
5	9	1.08	1.17	6.75	−429	−672	5879	−3132	−4904	42,913	−62	−97	850
10	20	2.4	2.6	15	−954	−1493	13,065	−6960	−10,897	95,362	−138	−216	1888
15	30	3.6	3.9	22.5	−1430	−2239	19,597	−10,440	−16,346	143,043	−207	−324	2833
20	35	4.2	4.55	26.25	−1669	−2613	22,863	−12,180	−19,070	166,883	−241	−378	3305
25	38	4.56	4.94	28.5	−1812	−2837	24,823	−13,224	−20,705	181,187	−262	−410	3588
30	39	4.68	5.07	29.25	−1859	−2911	25,476	−13,572	−21,250	185,956	−269	−421	3682
45	41	4.92	5.33	30.75	−1955	−3060	26,782	−14,268	−22,339	195,492	−283	−442	3871

and

Table A7. Adoption rate of influence of radio, crop productivities of adopted numbers, and number of people able to be fed under more rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed by the Produce Added
Radinf	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
5	9	1.71	6.39	0.9	327	10,106	−137	2389	73,764	−1003	47	1461	−20
10	20	3.8	14.2	2	727	22,457	−305	5309	163,919	−2230	105	3246	−44
15	30	5.7	21.3	3	1091	33,685	−458	7963	245,879	−3345	158	4869	−66
20	35	6.65	24.85	3.5	1273	39,299	−535	9291	286,859	−3902	184	5680	−77
25	38	7.22	26.98	3.8	1382	42,668	−580	10,087	311,447	−4236	200	6167	−84
30	39	7.41	27.69	3.9	1418	43,791	−596	10,352	319,643	−4348	205	6330	−86
45	41	7.79	29.11	4.1	1491	46,036	−626	10,883	336,035	−4571	216	6654	−91

) (Figure 9). Farmers in low-density areas and areas with low literacy rates could receive seasonal climate forecasts through radio [39,92]. A total of 63% of farmers of the district of Murewa identified radio as an important and reliable source of weather information in Zimbabwe [40]. Radio broadcasting is considered a dissemination channel for farmers’ awareness of weather forecast information. Nearly 66% farmers reported they received seasonal forecast information by listening to the radio and considered the information in their decision making in a study by Zongo et al. [93]. Another study shows farmers’ knowledge about climate information improved in one village of the Congo through radio programs [94].

3.4. Farmers’ Interactions with Agriculture Extension Workers

Improving the influence level of extension workers up to 30% would result in 27% more adoption rates. Double or triple the influence would improve adoption rates by 34% to 40% per season. This implies that 530 smallholder farmers may use scientific weather forecast information [64].

This increase in adoption rates could bring about 200,000 to 350,000 kg of maize into the community (

Table A8. Adoption rate of farmers’ interactions with agriculture extension workers, crop productivities of adopted numbers, and number of people can be fed under less rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed by the Produce Added
Ext inf	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
5	5	0.6	0.65	3.75	−238	−373	3266	−1740	−2724	23,840	−34	−54	472
10	20	2.4	2.6	15	−954	−1493	13,065	−6960	−10,897	95,362	−138	−216	1888
15	30	3.6	3.9	22.5	−1430	−2239	19,597	−10,440	−16,346	143,043	−207	−324	2833
20	38	4.56	4.94	28.5	−1812	−2837	24,823	−13,224	−20,705	181,187	−262	−410	3588
25	38	4.56	4.94	28.5	−1812	−2837	24,823	−13,224	−20,705	181,187	−262	−410	3588
30	40	4.8	5.2	30	−1907	−2986	26,129	−13,920	−21,795	190,724	−276	−432	3777
45	42	5.04	5.46	31.5	−2002	−3135	27,435	−14,616	−22,884	200,260	−289	−453	3966

and

Table A9. Adoption rate of farmers’ interactions with agriculture extension workers, crop productivities of adopted numbers, and number of people able to be fed under more rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed by the Produce Added
Ext inf	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
5	5	0.95	3.55	0.5	182	5614	−76	1327	40,980	−557	26	811	−11
10	20	3.8	14.2	2	727	22,457	−305	5309	163,919	−2230	105	3246	−44
15	30	5.7	21.3	3	1091	33,685	−458	7963	245,879	−3345	158	4869	−66
20	38	7.22	26.98	3.8	1382	42,668	−580	10,087	311,447	−4236	200	6167	−84
25	38	7.22	26.98	3.8	1382	42,668	−580	10,087	311,447	−4236	200	6167	−84
30	40	7.6	28.4	4	1455	44,914	−611	10,618	327,839	−4459	210	6492	−88
45	42	7.98	29.82	4.2	1527	47,159	−641	11,149	344,231	−4682	221	6816	−93

) (Figure 10). Local government can reach this higher adoption and yield improvement with a 30% improvement of extension service influence and capacity in the Kebele [64]. Fadare et al. [95] studied the factors influencing the adopting decisions of maize farmers in Nigeria. They found 87.4% do not have access to extension service, and the parameter estimates of the model to identify the factors influencing farmers’ decisions show acquiring access to extension service would increase probability of adoption by 0.47 compared to non-adopter counterparts. Maoba [96] studied farmers’ perceptions of agricultural extension service in Gauteng, South Africa. The study used the Likert-type scale method to assess the effectiveness of extension methods. Training and demonstration methods (mean value 3.6, SD = 0.5) were perceived to be highly effective by farmers compared to farmer days, office calls, and print materials. Farmers also found regular visits of extension specialists significant, such as biweekly and monthly compared to occasional and irregular. Extension services also can explain and interpret seasonal forecasts correctly for farmers and help make proper decisions on farming activities [97]. It has also been reported that inadequate support of extension services is also one of the major reasons for farmers’ refusal and use of seasonal forecast information. An amount of 75% of farmers are unaware of seasonal forecasts in West Africa. A total of 90% of survey respondents in 13 communities said they would definitely use climate forecast information if it were available to them in Burkino Faso [98]. Thus, agriculture extension workers are a significant influencer of farmers’ adoption of weather forecast information. Government agency programs on human capacity programs would significantly improve this staple food of rural populations.

3.5. Impact of Forecast Accuracy on Agricultural Productivity

Forecast accuracy improvement shows positive adoption and yield productivity in the Kebele. A 5% improvement of weather forecast accuracy (from 65% to 70%) shows a 23% increased adoption rate. A 15% improvement (from 65% to 80%) would increase adoption by 37% [64].

Improving forecast accuracy to 70% would bring 175,000 to 303,000 kg more yield. However, forecast accuracy failure significantly drops adoption rates and reduces yield productivity by about 50% (

Table A10. Adoption rate of weather forecast accuracy, crop productivities of adopted numbers, and number of people able to be fed under less rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed for the Produce Added
Accur	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
30	0	0	0	0	0	0	0	0	0	0	0	0	0
40	0	0	0	0	0	0	0	0	0	0	0	0	0
50	1	0.12	0.13	0.75	−48	−75	653	−348	−545	4768	−7	−11	94
55	10	1.2	1.3	7.5	−477	−746	6532	−3480	−5449	47,681	−69	−108	944
60	20	2.4	2.6	15	−954	−1493	13,065	−6960	−10,897	95,362	−138	−216	1888
65	30	3.6	3.9	22.5	−1430	−2239	19,597	−10,440	−16,346	143,043	−207	−324	2833
70	37	4.44	4.81	27.75	−1764	−2762	24,169	−12,876	−20,160	176,419	−255	−399	3493
80	41	4.92	5.33	30.75	−1955	−3060	26,782	−14,268	−22,339	195,492	−283	−442	3871
90	43	5.16	5.59	32.25	−2050	−3210	28,089	−14,964	−23,429	205,028	−296	−464	4060

and

Table A11. Adoption rate of weather forecast accuracy, crop productivities of adopted numbers, and number of people able to be fed under more rain conditions.

		Baseline Survey			All Sampled Farmers Maize Produce, kg, Added			All Farmers in the Kebele Maize Produce, kg, Added			Number of People Able to be Fed for the Produce Added
Accur	# Adoptions	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later	Plant Normal	Plant Earlier	Plant Later
30	0	0	0	0	0	0	0	0	0	0	0	0	0
40	0	0	0	0	0	0	0	0	0	0	0	0	0
50	1	0.19	0.71	0.1	36	1123	−15	265	8196	−111	5	162	−2
55	10	1.9	7.1	1	364	11,228	−153	2654	81,960	−1115	53	1623	−22
60	20	3.8	14.2	2	727	22,457	−305	5309	163,919	−2230	105	3246	−44
65	30	5.7	21.3	3	1091	33,685	−458	7963	245,879	−3345	158	4869	−66
70	37	7.03	26.27	3.7	1346	41,545	−565	9822	303,251	−4125	194	6005	−82
80	41	7.79	29.11	4.1	1491	46,036	−626	10,883	336,035	−4571	216	6654	−91
90	43	8.17	30.53	4.3	1564	48,282	−657	11,414	352,427	−4794	226	6979	−95

) (Figure 11). Thus, failure of forecast accuracy poses some cost risks to smallholder farmers. Forecast accuracy can be a determinant of the impact of seasonal forecast information in the community [99].

The results of our study can be applied to different countries in Africa to help smallholder farmers improve their adaptive capacity to climate change. A key strength of the ABM in this study is in linking farmer management strategies and agriculture productivity to develop a complex human–environment agriculture model. The distinctive features of ABM enable the study of the emergence of collective responses due to the use of seasonal forecast information and provide new insights into agriculture management. Our model is likely to be applicable in developing countries where rainfed agriculture is common such as Zimbabwe, Benin, Kenya, and Nigeria. Cane et al. [100] showed maize yields were correlated to seasonal rainfalls. Many farmers showed interest in receiving weather forecast information from radio broadcasts in Benin, West Africa [101]. Farmers in Benin found that access to extension services helped them to understand the usefulness of information [39]. Weather forecast through mobile phone technology is highly appreciated in Kenya [102]. The agricultural production system is exclusively rainfed in Nigeria; hence, the farmers are vulnerable to seasonal variables [103].

4. Conclusions

The adoption of seasonal forecasts helps farmers in farm management decisions to obtain the maximum benefits of agriculture productivity. Farmers can optimize their use of fertilizers, chemicals, seed amounts used, and other inputs, as well as sowing timings and tillage frequencies. Although weather forecast information has proven to help farmers manage on-farm risks, the use and adoption of forecast information still remains low in Ethiopia. In this study, we assessed the impact of weather forecast information adoptions and productivity improvement in Rim Kebele, Amhara Region. We coupled ABM with crop models to simulate crop productivity responses to different farmer management strategies under less and more rainfall conditions. Survey analysis indicated that maize is the foremost cultivated crop in the Kebele. Rim Kebele maize productivity of 3500 kg/ha is higher than the average national productivity of 3000 kg/ha but lower than the world average yield of 4500 kg/ha.

The ABM part of the model identifies the different information flow pathways where farmers communicate and determines the weather forecast adoption rates. There are four leading influencers in information communication in the Kebele: nearby farmers, other farmers in the Kebele, agriculture extension workers, and radio. The most direct way to disseminate weather forecast information to farmers is through agriculture extension agents and cooperation with them. Outreach to the local authorities such as Kebele leaders, the development of weather bulletins, and broadcasting through radio are other means of disseminating weather forecast information.

The crop model simulation uses generated rainfall pattern data for less and more rainfall conditions, soil data, farmers’ planting time preferences, seed amounts, tillage, and other inputs such as fertilizer use and herbicide use. We calibrated the model based on survey productivity data. Our results show that farmers’ decisions based on weather forecast information improve crop productivities from 17% to 30% under less and more rainfall conditions, respectively. Farmers’ inter- and intra-village interactions, improved farmer connections, radio, and agriculture extension service influences increased adoption rates and brought additional crop yields into the Kebele compared to non-forecast users. Weather forecast adopters may bring an additional 170,000 kg to 350,000 kg of maize into the Kebele community. This would be sufficient to feed 3500–6500 more people. Improving forecast accuracy has also shown positive yield supplements.

This study used ABM to link farmers’ management strategies and crop productivity to understand the complex farmer–environment system. ABM helps to study collective emergence behavior of farmers in response to different agricultural management policies. Our results can be applied in other developing countries where rainfed agriculture systems are common.

Our study’s limitations include the fact that we assumed farmers’ decisions on planting date, using fertilizer, herbicides, and tillage frequencies. We did not look at the other factors such as seed types, seed densities, seed price fluctuations, farmers’ other crop types and his/her cultivation in larger or smaller plot areas. Our analysis and results are based on survey data collected during 2019. Future work could expand this research over additional years and compare the farmers’ uptake of weather forecast information and its impact on the crop productivity in the area. Our findings help local government policy makers understand the impact of the use and dissemination of forecast information in the Kebele and its impact on agricultural productivity. Results of this study can be used to develop agricultural programs for farmers and other users of weather forecast information.