Using machine learning and smartphone imagery to estimate and manage weather-related risks to smallholder agriculture

• Supervisors: Dr Tim Foster (MACE)
  Dr Angela Harris (SEED)
  Dr Luis Garcia-Carreras (SEES)
  
  
• External Supervisors: Dr Berber Kramer (IFPRI)
  
  
• Contact:

  Dr Tim Foster, timothy.foster@manchester.ac.uk

• CASE Partner: N/A

Application deadline: 30 May 2018

Extreme weather events, such as heat waves, droughts, and floods, have severe negative impacts on agricultural productivity and livelihoods of smallholder farmers across the developing world. Expectation of crop failures discourages farmers from making productivity-enhancing investments, trapping smallholders in low-risk low-return agriculture and exacerbating vulnerability to future climate change (Vermeulen et al., 2012). 

One solution to this challenge is agricultural insurance. Crop insurance provides farmers with financial protection against crop losses caused by extreme weather, and has been widely advocated as a tool to help smallholders to escape poverty traps and to unlock investments in climate-smart agriculture (Barnett et al., 2008). However, to date, there have been few examples of successful crop insurance schemes in the developing world. A major reason for this lack of success is the difficulty in verifying crop losses in heterogeneous smallholder systems due to either: (i) the time-consuming verification of yield losses for traditional indemnity insurance, or (ii) the difficulty designing reliable proxies (e.g. weather indexes) for yield losses in the case of parametric insurance (Cole et al., 2013; Carter et al., 2017). 

Advances in high-resolution satellite imagery and smartphone ownership offer new opportunities for improving loss verification in smallholder agricultural systems. Recent work by our end-user partner, the International Food Policy Research Institute (IFPRI), has demonstrated the ability to use smartphone pictures of crops for manual detection of growth status and damage from extreme weather (Kramer et al., 2017). This approach has the potential to enhance the reliability and affordability of crop insurance payouts, and increase farmers’ trust in these products. However, implementing picture-based crop insurance at scale will require new tools for detected automatically crop features/damage without the need for manual image assessment by expert agronomists. 

Machine learning and artificial intelligence techniques could help to unlock the potential of big data for managing climate risks to agriculture (Cai et al., 2018; Kamilaris et al., 2018). For example, deep machine learning approaches, such as artificial neural networks or decision trees, could be trained to automate extraction of features relevant to yield estimation such as crop type, growth stage, or visible damage caused by extreme weather. While manual extraction of these features from satellite or in-situ imagery is time consuming, automated feature extraction in near real-time could be linked with crop models and forecast data to provide in-season loss projections to speed up insurance payouts, or to support individually-tailored advisory services (e.g. for irrigation scheduling) to enable farmers to take precautionary measures to minimise weather-related crop damage.

The overall aim of this project will be to assess the ability to use high-resolution smartphone and satellite imagery to verify, predict, and mitigate crop yield losses in smallholder agricultural systems through use of machine learning and AI methods. Improving agricultural productivity is central to multiple UN Sustainable Development Goals (SDG’s), and is a key stated objective of the UK Industrial Strategy’s theme on AI & Data-Driven Economy. Key research questions and topics that will be explored could include: 

(1)   Monitoring crops and management practices: Can machine-learning methods reproduce manual expert assessments of crop condition and visible management practices based on crop imagery, and what/how much data is needed for reliable estimation?

(2)   Yield estimation: Does information extracted from crop imagery increase the accuracy of yield predictions from crop growth models, and what information is most valuable for constraining estimates of weather-yield relationships?

(3)   Improving resilience: Can automated image analysis be used to predict yield losses in advance of harvest, and at what lead-time? Can this be combined with agro-advisories to enhance overall resilience to weather extremes? 

The project would be ideally suited to a student with a strong quantitative background in computer science, engineering, or physical sciences. Prior knowledge of machine-learning techniques is not a pre-requisite. Under the guidance of the supervisory team, you will gain a wide breath of cross-disciplinary training in remote sensing, image analysis, crop-climate modeling, and machine learning techniques. You will also work with partners at IFPRI (Dr Berber Kramer), who will support the project by providing access to in-situ crop imagery and field data (e.g. crop yields) from picture-based insurance pilots in South Asia and Africa where the proposed methods will be tested.

Barnett, B. J., Barrett, C. B., & Skees, J. R. (2008). Poverty traps and index-based risk transfer products. World Development, 36(10): 1766-1785. 

Cai, Y., et al. (2018). A high-performance and in-season classification system of field-level crop types using time-series Landsat data and a machine learning approach. Remote Sensing of Environment, 210: 35-47. 

Carter, M., de Janvry, A., Sadoulet, E., & Sarris, A. (2017). Index insurance for developing country agriculture: A reassessment. Annual Review of Resource Economics, 9(1): 421-438. 

Cole, S., et al. (2013). Barriers to household risk management: Evidence from India. American Economic Journal: Applied Economics, 5(1): 104-135. 

Kamilaris, A., & Prenafeta-Boldú, F. X. (2018). Deep learning in agriculture: A survey. Computers and Electronics in Agriculture, 147: 70-90. 

Kramer, B., et al. (2017). Picture-based crop insurance: Is it feasible? Using farmers’ smartphone pictures to minimize the costs of loss verification. IFPRI Project Note 01. 

Vermeulen, S. J., et al. (2012). Options for support to agriculture and food security under climate change. Environmental Science & Policy 15(1): 136-144.

Apply Now