Agricultural Statistics, Data Science and a Crop Map

May 21, 2021 by No Comments | Category Land use

A mosaic picture of the fields of an anonymous selected area colour coded by predicted crop type

Mosaic of crop types

As an analyst with a passion for maps and data science, I am excited about our new data science crop map of agricultural land-use. It helps to improve the accuracy of our statistics which I mentioned in my last blog post. But first let me highlight some other releases, as it has been a busy month for the team in agricultural statistics.

Recently Published Agricultural Statistics

In April the ‘Farm Business Income’ report was published. It gives a rich source of data for business performance for 2019. Later this month we will put out more detailed tables by region for farm types like beef, cereal and mixed farms. ‘Total Income from Farming’ estimates, the third major release of the year, also came out this month. The focus of this is to estimate economic output from farming in 2020.

A new Scottish Crop Map

And now our analysts are adding a new and exciting ‘Scottish Crop Map’ to the list. We will be publishing this at 9:30 on the 27 May. The map will provide field level data on farm activity and land-use across the whole of Scotland for the first time. It will also be released under the UK Statistics Authority ‘Experimental Statistics’.

The Scottish Crop Map moves us away from using surveys, and instead uses a mixture of satellite radar images with artificial intelligence. To produce it, the data science team in the Scottish Government have developed a machine learning algorithm. And with thanks to our collaborators, we have been supported by Edinburgh University’s data and digital expertise at EDINA and the Joint Nature Conservation Committee (JNCC).

The technique teaches a super-computer to recognise the crops growing in a field from radar images. We then train it to recognise from 10,000 field inspections to identify crops growing in 400,000 fields across Scotland.

Get Involved

As this is the first attempt to produce agricultural statistics using this method, the team are still revising the methods against the real world. We are publishing this as ‘Experimental’ as the next steps will be to gather feedback. We are looking for comments from farmers to data specialists. I would encourage any member of the general public to provide feedback. This will help improve the techniques that are being tested and improve the accuracy of the results. You can provide corrections and feedback using the link on the map.

The focus of the data science crop map is firmly on getting feedback and sharing progress made so far. To support this ethos, the map will be published allowing people to interact with it. It is a highly visual map showing every field in Scotland. A taster of what it will look like at the most granular level is shown in the picture. The team will welcome feedback and corrections through the website where the map is published.

Open Source Code and Data Science

The code and machine learning algorithm which has been created to predict the crop growing in every field is in GitHub. It is already available as an ‘open source’ development. This is to allow specialists in data science, agronomy and anyone with a keen eye for code to download it and scrutinise the processes. Accompanying documents will help people navigate the methods being deployed. It also explores the reasons for choosing the methods and alternatives tested.

There will be access to the data behind the map. The data from the satellites is already available but we will provide the ‘cleaned’ data set alongside our code. We are also making the base map data available under license.

This is the first attempt of the map and our team are planning on providing yield estimates in future years.

Would you like a reminder when the Scottish Crop Map is published? If so, please leave a reply or email the team.

Tags: , , ,


Leave a comment