Take-Home (Spatial) Messages

What have we learned, what we'd like to tell the community and partners, and what's next? Here are some of our reflections from the CSI Meeting in Cali.
Ruben (CIAT DG) opens the CGIAR Big Data Convention on Sep 9, 2017

It’s been about two weeks since the epic 2017 CGIAR Platform for Big Data in Agriculture Convention. Before all the excitements and momentum fade, now would be a good time to write time some key take-home messages from CSI’s perspective. Coincidentally (and quite conveniently), I just received a list of questions from Brian (the Platform Leader, in case you haven’t met him yet) to help walk back to our memory lane.

What is cutting edge work and who is doing it?

Beyond all the cool things what we (CSI Community) already have been doing, I think this year’s convention very effectively highlighted the proven potential of using the exploding amount of high-resolution remote sensing imagery and machine learning in agriculture.

Almost effortlessly, George mapped the NDVI of Cali from the imagery taken the day before presentation, using NASA imagery on Google Earth Engine.

During the Convention plenary on Why Big Data in AgricultureGeorge Azzari from Stanford University showcased these applications applications on crop yield mapping, mapping field management practices, and crop type classifications. Several CGIAR Centers have been also working on this over the years, but George’s presentation raised the bar with the near-instantaneous processing time of near-real time imagery over large, continental-scale.

What do we want to tell the sector as a whole about good practice in applying geospatial science in agriculture?

Besides all the cool technological and data breakthrough, we also shared a bit of cautious tales. We are not quite there yet! As much as the potential, we all should be very clear on the limitations of using remote sensing. For example, even with the super-high resolution imagery taken from the latest satellites above sky, most of the satellites can’t do anything with clouds.

Cloud contamination, the single most terrifying nightmare for remote sensing people

Also, all the machine learning algorithm does not, and would not, diminish the importance of quality groundtruthing data. Especially in the areas with heterogeneous land use and farm management practices, we will need a lot of training data to make good sense of what you see in the remote sensing data. We can’t overstate the importance of good sampling framework, to properly (and efficiently) represent the spatial variability of field-measured data in the training dataset. Throughout the discussion sessions, several “Call to Action” items were identified, such as the global collection of standardized groundtruthing data and spectral signatures. This can’t be done by a single initiative or small group of people – so let’s partner!

What issues or complexities need to be worked through to make sure geospatial community moves along a path that will be of benefit to small farmers?

Within the CSI Community, to move along the so called impact pathway, we discussed that we should invest on ourselves, to training on computer programming for geospatial analysis (rather than fully rely on GIS software).

Crash Course: Programming for Geospatial Analysis (Robert Hijmans, UC Davis)
Crash Course: Getting Started with Programming for GIS (Robert Hijmans, UC Davis)

Many of our community members do some programming already, but not so many were using languages most widely used in the geospatial big data analytics, like R or Python. For this, we were able to organize a quick crash course on Getting Started with Programming for GIS by Robert Hijmans during the Convention. We will continue organize community-wide activities to help self-train on the use of these new data and tools.