Seeing is believing: Using Crop Pictures in Personalized
Advisory Services

Francisco Ceballos, Tim Foster, Koen Hufkens, Arun Jadhav, Samyuktha Kannan, and Berber Kramer

Policymakers in many low- and middle-income countries have tried to promote affordable, effective risk management instruments, such as crop insurance, to help protect smallholder farmers from losses induced by climatic risks. Thus far, however, the number of successful crop insurance schemes targeting smallholders has been limited, arguably due to three main reasons: (1) high monitoring and verification costs of traditional indemnitybased insurance and, to an extent, of area-yield index-based insurance; (2) poor trust and high basis risk (i.e., imperfect correlation between farmers’ actual losses and insurance payouts) among farmers, leading to low demand for weather- and area-yield index-based insurance; and (3) the fact that insurance products often act as a substitute for resilience-enhancing technologies that could help increase profitability and prevent crop damage, such as irrigation, drought- or heat-tolerant cultivars, and integrated pest and disease management.

This project note describesthe potential for personalized remote advisory services bundled with insurance, provided based on crop pictures from farmers’ fields, to promote the adoption of resilienceenhancing technologies. Specifically, we present findings from a study in which we tested the feasibility and impact of including advisories in a picture-based insurance (PBI) product in India.

PBI verifies claims of crop damage by using a series pictures of insured fields from sowing to harvest. These pictures are taken by farmers themselves with a tamper-proof smartphone application. Cellphone imagery gives insurers “eyes on the ground”, reducing monitoring costs and allowing them to provide affordable and highquality crop insurance to smallholder farmers. Previous project notes have highlighted the feasibility and sustainability of this approach: PBI minimizes basis risk and improves trust and tangibility by relying on farmers’ direct engagement with the product.

To further improve the product’s value proposition and enhance its usefulness as a climate change adaptation tool, we complemented PBI with personalized remote advisories based on real-time observations of crop conditions using farmers’ pictures of insured plots. In the winter of 2017/18, we developed and implemented this advisory service in eight districts of Haryana and Punjab, India.

We hypothesize that the bundled service empowers data-driven farming through three channels. First, the stream of on-the-ground pictures allows experts to target crop management recommendations to each farmer’s individual situation, improving the value and timeliness of the advice when compared to traditional advisory services. Second, the tangibility of pictures, together with farmers’ improved engagement, can increase ownership and take-up of the advice and create benefits even in years without insurance payouts. Third, the direct real-time observation of field conditions can allow insurers to gather valuable, previously unavailable monitoring data and to provide recommendations to help prevent crop damage— potentially lowering expected insurance payouts and thus insurance premiums. Through these three channels, this approach can improve insurers’ competitiveness, boost the sustainability of the insurance product, and create a business case for advisory services.

This project note explores each of these three channels using the results of a formative evaluation of the personalized advisory service. Specifically, we focus on three questions: (1) Does the advisory TAKE-AWAY MESSAGES • Identifying crop growth stages and improving yield predictions based on smartphone pictures of fields is feasible and scalable, allowing advisory services to be tailored to individual farmer needs. • Personalized remote advisories based on these pictures (picture-based advisories or PBA) improve farmers’ knowledge of recommended practices. • Farmers report that such advisories help them reduce risk, suggesting that bundling picture-based insurance (PBI) with PBA can improve their adaptive capacity and help to lower insurance premiums. • Bundling PBA with PBI also improves farmer engagement in, satisfaction with, and willingness to pay for the advisories, suggesting strong complementarities between these services. P 2 service allow experts to target messages to a farmer’s individual situation, thus increasing the value and timeliness of the advice? (2) Does the tangibility of pictures and improved farmer engagement increase ownership and take-up of the advice? (3) Does this system provide insurers with real-time monitoring data that allow them to provide recommendations to minimize risk?

Methods and data

During the winter of 2017/18, we developed and rigorously tested an advisory service using a cluster randomized trial in approximately 200 villages in selected districts of Haryana and Punjab, India (Fatehgarh, Ludhiana, and Patiala in Punjab; Fatehabad, Karnal, Panipat, Sirsa, and Yamunanagar in Haryana). We randomly assigned villages to one of three interventions: in 50 villages, we broadcasted conventional interactive voice response (IVR) and SMS messages (control group); in 75 villages, we added personalized, picture-based advisory messages (PBA treatment); in the remaining 75 villages, we provided PBI coverage on top of the IVR, SMS, and PBA messages (PBA + PBI treatment).

In the control group, farmers who had access to a smartphone and had landholdings of less than 15 acres were invited to participate in the survey. In the two treatment groups, farmers complying with the same criteria were also invited to download a free app, Wheatcam, and to register in the app to receive the PBA messages. Farmers could enroll one of their fields by registering one or more “sites” in the smartphone app, provided that the pictures taken at one site could capture approximately one acre of their field. During the registration process, farmers had to send in an initial geo-tagged and time-stamped picture for each of their registered sites. A total of 1,779 farmers from 141 villages in Haryana and Punjab registered on WheatCam; of these, 76 percent enrolled one site, 18 percent enrolled two sites, and the remaining 6 percent enrolled more than two sites.

Repeat pictures

After taking the initial registration picture, farmers were asked to take three pictures per week throughout the entire growing season. Pictures needed to be taken between 10am and 2pm (in order to keep lighting conditions constant) from exactly the same location as the initial picture and with the same view angle every time. To facilitate this, the smartphone app made use of geo-tags to check whether the repeat picture was taken at the same location as the initial picture. In addition to providing visual aidsin the form of a line to mark the horizon, the app also provided a “ghost” image (a partially transparent image of the initial picture; see Figure 1) on the smartphone screen when a farmer was taking a repeat picture, allowing the farmer to align static features in the landscape (such as distant trees or structures) with those in the initial picture. The ghost image approach helped ensure an almost identical view frame throughout the season. Farmers then uploaded valid pictures to an online server.

Advisories and loss moment

Figure 1: Ghost image of initial picture is shown when taking repeat picture

All farmers in the study received conventional advisory messages built on CABI’s Direct2Farm program, which had been implemented in the area one year earlier, together with generic weather advisories for the area. Content was recorded and broadcast as IVR messages to all phone numbers registered in the study database; in addition, these messages were also sent out as SMS messages.

Farmers in the treatment groups (PBA and PBA + PBI) received additional personalized PBA messages, through either the app or SMS, when they submitted a repeat picture or contacted agronomic experts through the app. Four local agronomists interpreted the uploaded images (including close-up pictures of the field, requested by the app when a farmer declared to had suffered damage) and sent out personalized advisories based on cues visible in the pictures and additional sources of information such as weather data and regional pest monitoring. For this purpose, they used an online platform linked to the smartphone application that allowed them to accept or reject individual farmer’s pictures (according to whether the farmer respected the stated picture-taking protocol), review the images for visible cues to prompt specific crop management recommendations, and push remote advisories (PBA messages) directly through the app to each farmer’s phone. In addition, at the end of the season, these experts assessed the level of visible damage at each site using the time lapse of pictures. Assessments were made individually, and the median percentage of damage across experts was used as the final damage measure for that site. When large disagreement existed among individual assessments, we used the percentage of damage reached by consensus during a joint review.

In total, we received and analyzed 9,923 pictures and broadcast IVR and SMS messages to 32,237 wheat producers. A subsample of 1,179 beneficiaries received 5,081 personalized PBA messages, targeted to their individual needs.


In January 2018, the 985 beneficiaries from randomly selected PBI villages received insurance to cover the sites they enrolled in WheatCam during the ongoing Rabi (winter) 2017/18 season against visible crop damage and extreme heat. Insurance was conditional on taking regular pictures of their plots using WheatCam. We told farmers that their insurance included coverage against damage visible in their pictures (PBI), as determined by experts during the loss assessments, as well as coverage against above-normal temperatures between January 21 and March 20, as measured at nearby weather stations. For farmers with more than 20 percent of assessed visible damage, we sent a damage report, including the pictures and the expert loss assessments, to HDFC Ergo General Insurance Company (henceforth HDFC), the project partner co-developing and underwriting the insurance product. HDFC then issued payments directly into farmers’ bank accounts.

For the 2017/18 season, the project used these expert assessments as a transparent, pragmatic solution to providing PBI coverage in the absence of automated tools to estimate damage from the time lapse of pictures. In the future, after additional training data is collected both for wheat and for other crops, we will develop image processing and machine-learning algorithms to automate the loss assessment process. Such automation will constitute an important public good to encourage the adoption of this approach at scale.

Evaluation data: Crop cutting and survey

At the end of the season, researchers visited 638 of the photographed plots to conduct crop cutting experiments for yield measurement. For each plot, the researchers sampled two different square meters that were visible in the pictures: one to the left and one to the right of the picture. The heads of the wheat plants falling inside these sampled square meters were threshed, the resulting grains were weighted, and the average weight from these two square meters was used to calculate yields per acre. We did not use these yield data as input in the loss assessments; the primary reason for collecting these data was to have an objective measure of yields, a critical step for assessing the validity of loss assessments from farmers’ own smartphone pictures.

For further validation and evaluation, we also interviewed a subsample of 529 households at endline (479 from the PBA and PBA + PBI villages and 50 farmers from the control group villages). We designed the endline survey to measure farmers’ knowledge and adoption of best practices (e.g., input use), damage suffered during the season, farmers’ satisfaction with the insurance product and advisory service, and farmers’ willingness to pay for PBA, PBI, and PBA + PBI.

Automated image processing

Finally, we explored to what extent crop pictures can be processed using automated procedures to monitor crop growth stages. For this component, described in more detail in Hufkens et al. (2018), we extracted greenness indices from smartphone pictures of wheat crops (taken during the 2016/17 season, in which we followed similar insurance procedures but did not provide advisories). We also added crop growth stage labels, indicating whether the crop in the picture was in the tillering phase (soil visible), stem extension phase (no soil or wheat ears visible), or heading and ripening phase (wheat ears visible). We used these data to analyze whether the greenness indices derived from a stream of pictures for a given plot were predictive of the growth stage on that plot.


1. Does the advisory service allow experts to target messages to a farmer’s individual situation, increasing the value and timeliness of the advice?

Different growth stages imply exposure to different types of risk, and thus call for specific advisory messages. Real-time observations of crop growth stages based on farmers’ pictures could thus be used to provide more targeted messages and recommendations for weather- and non-weather-related risk management. As an initial step to gauge the potential for this, we used related research to analyze whether crop pictures could be used to infer important wheat crop growth stages (Hufkens et al. 2018). We found that greenness index curves derived from crop picturestaken throughout the entire season can indeed predict the transition from tillering to stem extension and from stem extension to heading and ripening with substantial accuracy (Figure 2). Moreover, the predictive power of using greenness indices derived from ground pictures was found to be higher than that of conventional satellite vegetation indices (Hufkens et al. 2018).

Figure 2: Growth stage and greenness over time

A separate approach tested whether integrating smartphone pictures into a convolutional neural network would improve yield prediction. These analyses were conducted by BKC WeatherSys, one of India’s first private sector meteorology and environmental technology companies and the first private sector entity in India to run numerical weather prediction models. Yields were indeed predicted with higher accuracy when smartphone pictures were included in their existing agronomic software that generates automated advisories (which presently combines crop models, weather data, and satellite imagery). Moving forward, we will partner with BKC WeatherSys to further explore the use of smartphone pictures to strengthen their existing advisory service.

2. Does PBA increase ownership and take-up of the advice?

We first tested farmers’ knowledge of practices, recommended through both the generic IVR/SMS and the PBA messages, at endline. Farmers in both the control group and the two treatment groups completed a knowledge test with five items. The personalized advisory messages increased farmers’ knowledge of these recommended agricultural practices by a statistically significant 78 percent, from an average of 0.64 correct answers in the control group to an average of 1.14 correct answers in the treatment groups (Figure 3). Interestingly, the difference in knowledge remained when we focused on the subset of questions related to information provided through both IVR and PBA (second set of bars) and through the PBA service alone (third set of bars).

This suggests that farmers incorporate content better through the PBA approach even when this content does not differ in essence from what is provided through the generic IVR messages, arguably related to a sense of increased ownership with this approach. Because treatment was randomized, we can attribute these effects to the PBA messages and conclude that personalized messages had stronger impacts on knowledge than generic IVR/SMS messages alone.

Figure 3:PBA improved knowledge of recommended practices

Furthermore, we analyzed the effect of receiving PBA on the adoption of herbicides, pesticides, and fungicides recommended through these messages (the endline survey did not include information on other recommended practices). Consistent with other studies on the impacts of agricultural advisory services (Aker, Gosh, and Burrell, 2016; Cole and Fernando, 2016), we found—despite positive effects on knowledge—no strong short-term effects on adoption of most management practices (Figure 4). The PBA messages did not have an effect on pesticide or weedicide use, although they resulted in a small but statistically significant reduction in recommended weedicide use.

Since farmers in this treatment group were only slightly more likely to suffer damage from weeding (with the difference between groups being not statistically significant), this suggests that the more personalized and targeted advice allowed farmers to economize on unneeded herbicide usage.

Figure 4: PBA did not have a strong effect on adoption of recommended practices

3. Does this system provide insurers with real-time monitoring data and allow them to provide recommendations to minimize risk?

In the long run, one could envision the advisories having stronger impacts on management practices, productivity, and profitability. In that case, the advisories, by generating direct real-time observations of field conditions, could allow not only advisory service providers but also insurers to gather valuable, previously unavailable monitoring data and to provide recommendations to potentially help prevent crop damage, thus lowering expected insurance payouts and insurance premiums.

To test this hypothesis, we asked farmers how recommendations received from alternative advisory sources had helped them minimize their crop risk. Even though most participants reported the PBA and IVR advisories as being helpful in minimizing risk, farmers actually receiving PBA seemed to differentially recognize the advantage of the project’s advisory sources over other regular sources—e.g., radio, TV, agri-dealer—of agricultural advice (Figure 5).

Figure 5:Participants recognize PBA helping minimize risk better than other sources

Although this is a subjective measure, it is encouraging to see that participants strongly valued the PBA messages in this regard. This suggests that the PBA messages were a more effective way to provide recommendations on agricultural risk management and could in the future be integrated into insurance products to lower expected payouts and thus premiums. From an advisory point of view, we also found strong complementarities of bundling PBA with PBI. Figure 6 shows that engagement in the PBA service—measured as the number of pictures submitted and farmers’ satisfaction—was significantly higher when bundled with PBI. Figure 7 provides further evidence of complementarities by comparing farmers’ willingness to pay for PBA and PBI when offered in isolation and when offered as a bundle, as measured during the endline survey. While willingness to pay for PBA alone was negligible, respondents were willing to pay an extra Rs. 125.9, equivalent to 8.7 percent of the insurance premium, when advisories were embedded in the PBI product.

Figure 6: Providing PBI improves farmer engagement and satisfaction with advisory service


We developed, implemented, and evaluated an innovative personalized advisory service that complements picture-based insurance (PBI), an easy-to-understand low-cost insurance product for visible crop damage. We sent personalized agricultural advice based on real-time observations of crop conditions, from sowing to harvest, from farmers’ pictures of their insured plots. Such a service can empower data-driven farming through three channels: experts can target messages to a farmer’s individual situation, thus increasing the value and timeliness of the advice; the tangibility of pictures increases ownership and take-up of the advice; and the service allows insurers to gather more monitoring data and provide recommendations on how to minimize risk, thus lowering expected insurance payouts.

Figure 7: Higher willingness to pay for bundled product reveals synergies between insurance and advisories

We find that greenness indices derived from crop imagery can predict the onset of growth stages during which crops are more vulnerable to weather risk, outperforming satellite vegetation indices. Incorporating smartphone images improves the predictive power of existing automated advisory models, allowing for the provision of messages targeted to a farmer’s individual situation at scale. Survey data shows that advisory messages increased farmers’ knowledge on best agricultural practices, suggesting that the tangibility of pictures and the personalization of the advice increase ownership and can potentially serve to encourage take-up of the recommended practices. Finally, we find strong complementarities between PBA and PBI; farmers report that the PBA messages helped them reduce risk (potentially allowing for a reduction in insurance premiums), while farmer engagement in and willingness to pay for the PBA service significantly improved when bundled with PBI. This indicates strong synergies between insurance and advisories.

All in all, the findings from our formative evaluation point to the substantial advantages of combining picture-based insurance with advisory services. The synergies between insurance and advisory services could potentially enable insurance products to act as a complement to climate-smart resilience technologies, if advisories encourage the adoption of these technologies. Personalized advisories based on series of smartphone pictures taken by farmers can improve the effectiveness and encourage the uptake of agronomic recommendations, reducing agricultural risk and improving farmers’ adaptive capacity. Moreover, bundling PBI with advisories can increase the monitoring capacity of insurance providers and induce lower premiums in the long term, further adding to the attractiveness of the product.



Aker, J., I. Ghosh, and J. Burrell. 2016. “The Promise (and Pitfalls) of ICT for Agriculture Initiatives.” Agricultural Economics 47: 35–48. Cole, S.A., and A. Fernando. 2016. “‘Mobile’izing Agricultural Advice: Technology Adoption, Diffusion and Sustainability.” Harvard Business School Finance Working Paper No. 13-047. Hufkens, K., E. K. Melaas, M. L. Mann, T. Foster, F. Ceballos, M. Robles, and B. Kramer. 2018. “Monitoring Crop Phenology Using a Smartphone Based Near-Surface Remote Sensing Approach,” under revision.