Sunday, August 7, 2016

Challenges with Precision Agriculture: Finding the Balance Between Big Data and Local Conditions

Precision Agriculture (PA) is playing a major role in modernizing agriculture. PA usually means some sort of automated data collection (usually using sensors), often followed by an analysis (usually referred to as data mining) that rely on the historical data from multiple sites/farms. This analysis serves two goals: it provides manufacturers with better insight into their products, and it is also translated into recommendations. There is increasing need for farms to be as productive as possible, while also minimizing and mitigating any environmental impact of operations. Data mining, however, usually implies averages and correlations. Can these serve the goals stated above? There is a long history to prove the value of large-scale data mining, but it comes with a challenge. Although intended to do so, PA does not meet all the data needs of individual farms. There is a gap between multi-site data mining, on one hand, and local (farm-scale) data needs, on the other. Farming is much more than large-scale averages and correlations. Experienced farmers often have a “feel” for their land, which is difficult to translate into hard data that’s specific to a local site. It is said that “all politics is local”, and so is farming. Thus, data must be acquired and recommendations must be made based off of local farm scale data. Ideally, PA and local data can be coupled and complement each other in order to produce site-specific recommendations.

The global trends identified from data mining are important and helpful, but they neither intend nor are capable of addressing all the farm-scale challenges.  This is a well-known challenge in all large-sample statistics (from studies of human populations to spatial and environmental data): large-scale trends do not provide local-scale answers. For example in medicine, large scale pharmaceutical trials are obviously very important when considering population health, but they won’t tell you about the response of a particular individual to the proposed treatment. As a patient, you obviously care about the local-scale response, not just population averages.  Similarly, fish do not die of large-scale averages.

There is more to farming than a sensor in some soil. Farmers have their own ideas about what data to collect, where to collect it, and how to figure out which data is most useful.  This is true in many branches of agriculture. We can find an example for this challenge in an article by Noel Magnin, an agriculture expert, who commented on a well-known behavior in vineyards (LinkedIn, February 9, 2016): “Quality in wine grapes is due to some water constraint during the last steps of fruit maturation associated among other things with secondary metabolites production. Grape quality leads to wine quality and the best vintages occurred when water constraint is present and obviously when the maturation stages just before fruit collection have seen little to no rain. However, there are within vineyards locations where fruit quality never reaches a level high enough to result in high quality wine”. So, one can look at average crop quality parameters, trying to provide some general fixes, but it is the problem locations, where the soil becomes saturated at critical times, that would make the difference between poor, medium, and high quality grapes. You need to know where the trouble spots are, and large-scale averages won’t tell you that.   

Similarly, when managing cattle grazing operations, one can use some industry-based averages about grazing times and non-grazing intervals, but it would be more beneficial to modify these averages to reflect local conditions. Data mining provides good prior knowledge, but that prior knowledge must be updated by local conditions. This means reconciling between prior knowledge (the averages) on one hand, and site-specific evidence on the other. To do that, farmers need to explore the history of their farming operations in pictures, notes, sensor data, lab data and more, all dated and geo-referenced - and they should be able to explore that data with ease. This is, of course, not a new idea. What’s missing, however, are the tools that will allow farmers to do that.  
Let’s take a look at a few examples.

1. Data at your fingertips



What we have here is an example on how data could be organized in myObservatory, a web-based information management and analysis systems. The blue areas are hand-drawn shapes representing parcels/blocks/paddocks of particular interest. Each of these shapes acts as a container for all related data. This could include pictures, reports, notes, sensor data, lab reports, etc.   This system provides seamless connectivity between desktop and smartphones. So, for example, pictures can be taken using smartphones, and these pictures are automatically uploaded onto the myObservatory platform and automatically linked to the coordinate where it was taken. Data could be imported from external data providers for added insight, or data can be fed in by third parties (e.g. from partnered labs offering soil analysis lab services). Once data has been loaded or collected, it may be analyzed with statistical analysis and charting tools, geospatial analysis tools, or shared with selected stakeholders.

All Data are access-controlled. The project or site admin can assign users an appropriate level of access, anywhere from public view to adding data only, all the way to read/write privileges for any or all data and adding new users.  

 2Timeline



This example (courtesy Peter Traverse, Innisfree Farms, and myObservatory) shows time lapse photos showing one of the grazing areas in the farm. Want to know how long your herd grazed here? And how long it took for recovery to occur? Here it is, ready and available at your fingertips. Want to add notes? Want to link these images with lab data or with data from your groundwater wells or rainfall data? You can do it with a click of a button     

3. Story Viewer

The third example (courtesy Peter Traverse, Innisfree Farms, and myObservatory) tracks the movements of the herd. Each of the pictures was taken using a smartphone, which automatically geo-tagged and dated them. Once within transmission range, the picture is uploaded and stored in the farm’s database, ready and available for analysis. Notes can be added and auxiliary files may be attached.  You do not need a full-time photographer to take these pictures. You can have all your staff taking pictures with their smartphones, and then all these pictures would be automatically assembled and organized by myObservatory. All these pictures could be easily accessible and searchable by date and location.

In conclusion, farming is local, and farmers need a platform that will allow them to collect and explore their data with ease.  With Story Viewer, Timeline, and with easy access to data, you are ready to explore your data and make the decisions most suitable for your farm. For morre information, visit our website at my-observatory.com