Citizen science and remote sensing for crop yield gap analysis

The world population is anticipated to be around 9.1 billion in 2050 and the challenge is how to feed this huge number of people without affecting natural ecosystems. Different approaches have been proposed and closing the ‘yield gap’ on currently available agricultural lands is one of them. The concept of ‘yield gap’ is based on production ecological principles and can be estimated as the difference between a benchmark (e.g. climatic potential or water-limited yield) and the actual yield. Yield gap analysis can be performed at different scales: from field to global level. Of particular importance is estimating the yield gap and revealing the underlying explanatory factors contributing to it. As decisions are made by farmers, farm level yield gap analysis specifically contributes to better understanding, and provides entry points to increased production levels in specific farming systems. A major challenge for this type of analysis is the high data standards required which typically refer to (a) large sample size, (b) fine resolution and (c) great level of detail. Clearly, obtaining information about biophysical characteristics and crop and farm management for individual agricultural activities within a farm, as well as farm and farmer’s characteristics and socio-economic conditions for a large number of farms is costly and time-consuming. Nowadays, the proliferation of different types of mobile phones (e.g., smartphones) equipped with sensors (e.g., GPS, camera) makes it possible to implement effective and low-cost “bottom-up” data collection approaches such as citizen science. Using these innovative methodologies facilitate the collection of relatively large amounts of information directly from local communities. Moreover, other data collection methods such as remote sensing can provide data (e.g., on actual crop yield) for yield gap analysis. The main objective of this thesis, therefore, was to investigate the applicability of innovative data collection approaches such as crowdsourcing and remote sensing to support the assessment and monitoring of crop yield gaps. To address the main objective, the following research questions were formulated: 1) What are the main factors causing the yield gaps at the global, regional and crop level? 2) How could data for yield gap explaining factors be collected with innovative “bottom-up” approaches? 3) What are motivations of farmers to participate in agricultural citizen science? 4) What determines smallholder farmers to use technologies (e.g., mobile SMS) for agricultural data collection? 5) How can synergy of crowdsourced data and remote sensing improve the estimation and explanation of yield variability? Chapter 2 assesses data availability and data collection approaches for yield gap analysis and provides a summary of yield gap explaining factors at the global, regional and crop level, identified by previous studies. For this purpose, a review of yield gap studies (50 agronomic-based peer-reviewed articles) was performed to identify the most commonly considered and explaining factors of the yield gap. Using the review, we show that management and edaphic factors are more often considered to explain the yield gap compared to farm(er) characteristics and socio-economic factors. However, when considered, both farm(er) characteristics and socio-economic factors often explain the yield gap. Furthermore, within group comparison shows that fertilization and soil fertility factors are the most often considered management and edaphic groups. In the fertilization group, factors related to quantity (e.g., N fertilizer quantity) are more often considered compared to factors related to timing (e.g., N fertilizer timing). However, when considered, timing explained the yield gap more often. Finally, from the results at regional and crop level, it was evident that the relevance of factors depends on the location and crop, and that generalizations should not be made. Although the data included in yield gap analysis also depends on the objective, knowledge of explaining factors, and methods applied, data availability is a major limiting factor. Therefore, bottom-up data collection approaches (e.g., crowdsourcing) involving agricultural communities can provide alternatives to overcome this limitation and improve yield gap analysis. Chapter 3 explores the motivations of farmers to participate in citizen science. Building on motivational factors identified from previous citizen science studies, a questionnaire based methodology was developed which allowed the analysis of motivational factors and their relation to farmers’ characteristics. Using the developed questionnaire, semi-structured interviews were conducted with smallholder farmers in three countries (Ethiopia, Honduras and India). The results show that for Indian farmers a collectivistic type of motivation (i.e., contribute to scientific research) was more important than egoistic and altruistic motivations. For Ethiopian and Honduran farmers an egoistic intrinsic type of motivation (i.e., interest in sharing information) was most important. Moreover, the majority of the farmers in the three countries indicated that they would like to receive agronomic advice, capacity building and seed innovation as the main returns from the citizen science process. Country and education level were the two most important farmers’ characteristics that explained around 20% of the variation in farmers’ motivations. The results also show that motivations to participate in citizen science are different for smallholders in agriculture compared to other sectors. For example fun has appeared to be an important egoistic intrinsic factor to participate in other citizen science projects, the smallholder farmers involved in this research valued ‘passing free time’ the lowest. Chapter 4 investigates the factors that determine farmers to adopt mobile technology for agricultural data collection. To identify the factors, the unified theory of acceptance and use of technology (UTAUT2) model was employed and extended with additional constructs of trust, mastery-approach goals and personal innovativeness in information technology. As part of the research, we setup data collection platforms using open source applications (Frontline SMS and Ushahidi) and farmers provided their farm related information using SMS for two growing seasons. The sample for this research consisted of group of farmers involved in a mobile SMS experiment (n=110) and another group of farmers which was not involved in a mobile SMS experiment (n=110), in three regions of Ethiopia. The results from the structural equation modelling showed that performance expectancy, effort expectancy, price value and trust were the main factors that influence farmers to adopt mobile SMS technology for agricultural data collection. Among these factors, trust is the strongest predictor of farmer’s intention to adopt mobile SMS. This clearly indicates that in order to use the citizen science approach in the agricultural domain, establishing a trusted relationship with the smallholder farming community is crucial. Given that performance expectancy significantly predicted farmer’s behavioural intention to adopt mobile SMS, managers of agricultural citizen science projects need to ensure that using mobile SMS for agricultural data collection offers utilitarian benefits to the farmers. The importance of effort expectancy on farmer’s intention to adopt mobile SMS clearly indicates that mobile phone software developers need to develop easy to use mobile applications. Chapter 5 demonstrates the results of synergetic use of remote sensing and crowdsourcing for estimating and explaining crop yields at the field level. Sesame production on medium and large farms in Ethiopia was used as a case study. To evaluate the added value of the crowdsourcing approach to improve the prediction of sesame yield using remote sensing, two independent models based on the relationship between vegetation indices (VIs) and farmers reported yield were developed and compared. The first model was based on VI values extracted from all available remote sensing imagery acquired during the optimum growing period (hereafter optimum growing period VI). The second model was based on VI values extracted from remote sensing imagery acquired after sowing and before harvest dates per field (hereafter phenologically adjusted VI). To select the images acquired between sowing and harvesting dates per field, farmers crowdsourced crop phenology information was used. Results showed that vegetation indices derived based on farmers crowdsourced crop phenology information had a stronger relationship with sesame yield compared to vegetation indices derived based on the optimum growing period. This implies that using crowdsourced information related to crop phenology per field used to adjust the VIs, improved the performance of the model to predict sesame yield. Crowdsourcing was further used to identify the factors causing the yield variability within a field. According to the perception of farmers, overall soil fertility was the most important factor explaining the yield variability within a field, followed by high presence of weeds. Chapter 6 discusses the main findings of this thesis. It draws conclusions about the main research findings in each of the research questions addressed in the four main chapters. Finally, it discusses the necessary additional steps (e.g., data quality, sustainability) in a broader context that need to be considered to utilize the full potential of innovative data collection approaches for agricultural citizen science.