Creating accurate maps - the key to precision agriculture

Friday, May 1, 2015

A primer on how to make the maps for those year-to-year comparisons so essential to modern farming

by MIKE DUNCAN

The easiest way to make year-to-year comparisons in precision agriculture is by using maps of the variables of interest. The maps must be constructed so that they are the closest match to what is really in the field as possible.

To make maps from data in a farm field requires a set of calculations that I refer to as the precision agriculture pipeline (field data -> cleaning -> gridding -> map(s)…). For the sake of simplicity in the rest of this article, we will talk about yield maps, but everything we describe needs to be done to soils maps, chemistry maps, remote sensing maps and the like.

Precision agriculture maps are like everyday maps, except that they are typically composed of "cells" of some scale and represent things like yield across your fields. The size of the cell is often chosen to represent the size of a piece of equipment. At Niagara College, we typically use the size of the rake on the front of a combine as the scale. Twenty feet makes for a nice square cell size; however, other cell sizes can be used. (See Figure 1.)

Covering a 44-acre field in 20-feet x 20-feet cells takes about 5,400 cells that don't overlap. Each cell has GPS co-ordinates and a value. The map of cells looks like a grid covering the field. The bulk of precision agriculture calculations are made by comparing or calculating things from the maps. This article describes the steps in getting to the map.

All precision agriculture machinery thinks in terms of maps of cells. The technology that holds all of this together is GPS. GPS allows us to create seemingly imaginary "blocks" (a block is a group of cells with the same or similar values) within a farm field and have a GPS-enabled machine react to the edges and interior of the block as if it were real. (See Figure 2.)

The same idea applies to variable-rate machines. They can recognize when they are in a particular "cell" (or block) of the field and deliver a rate that is specified in a map for that cell. Variable-rate machines can adjust rates quickly to keep up with changes in the field. In order to program a variable-rate seeder or feeder, you have to give it a map, and it uses the boundaries of the cells covering the map to trigger the new rates so that it delivers the right amounts to each cell in the field.  

In the case of a combine, however, the yield monitor runs continuously and provides a stream of estimates of the yield. The monitor will sample the stream of yield estimates as fast as it can and assign a GPS coordinate to when and where the estimate was recorded. The sample then needs to be corrected for the time it took for the corn to be cut, processed in the turbine and then passed up a pipe into the hopper where the monitor is. The amount of correction is dependent on the speed of the combine.

In order to compare successive year's yield estimates from the combine, even if it has RTK auto-steer, the samples need to be mapped. Examining the co-ordinate assigned to samples from the same part of a field from successive years shows that the samples don't overlap, so we need to map them into cells in order to be able to compare them.

Prior to mapping/gridding, the data need to be "cleaned."  Bad data values can change the appearance and interpretation of maps. In our group's experience, even the best data often require a bit of touch up prior to use.

The cleaning process is usually pretty simple. Values that are way too high or way too low somehow end up in even the most carefully collected data by the most experienced collectors.  Too high and too low are relative terms defined with respect to the distribution of values that the combine collects.

When a distribution has an average of 133 bushels per acre with numerous highs up to 190 bushels an acre and numerous low values around 60 bushels an acre, a value like 265 bushels really stands out, as does a value of 20 bushels. (See Figure 3.) Cutting the unlikely values out actually involves the removal of less than 0.1 per cent of the data values, but results in a much clearer map of the variability. (See Figure 4.) In another article, we will look at some of the reasons these occur and how to clean them, but for now we will proceed to the next step.

Gridding or interpolation of data is a very mathematical process. The idea of gridding is very simple. You take yield values whose co-ordinates place them within or around a cell, and you use those values to assign a yield value to the cell. Most commercial software offers a number of interpolation methods, and each has its place.

In a subsequent article we will explore this idea more fully, but for now we will assume that we can use math to accurately calculate a value for each cell on the map such that it is representative of what is actually there. With high-density data sets like yield maps from combines, this is easier to do; with very sparse data sets like soil samples, this is much more difficult. However, once established, we can then assign a colour to ranges of values and the map ends up taking its characteristic mottled look. (See Figures 2, 3 and 4.)

A farmer with many years of data can perform these operations and end up with a stack of maps for each year that he or she has been collecting data. The stacks of maps are incredibly valuable to a farm business as they represent an exact picture of the farm's historical productivity over every square foot of each field.

After the pipeline has been used to create maps, then we can work on the products that the farmer can use to make more money;

1) In my last article in Better Farming (March 2015), we showed the Yield Probability Index (YPI) calculation. This calculation can also be performed using yield maps from a field over a number of years. This calculation needs only yield data, but produces a map of management zones.

2) The yield index (YI) is created by averaging yield maps of, for example, corn for the number of years collected. Most of the commercial software systems use the yield map as their basis for analyzing farm fields. If the farmer has seven years of corn data for a field, then add the maps up cell by cell and divide by seven to produce a map whose cells are now the average of the seven corresponding cells in the seven yield maps.

3) A landform class map is produced using a software tool called LandMapR to operate on a map of the surface elevation in a field (most RTK steering columns can collect this data).  

While not exciting or sexy, the creation of accurate maps is the key to precision agriculture.  Everything from management zones to remote sensing and soil sampling must be mapped in order to be useable. As we will demonstrate in later articles, all the math and calculations operate almost entirely on maps. BF

Dr. Mike R. Duncan, PhD, is Natural Sciences and Engineering Research Council of Canada, Industrial Research Chair for Colleges in Precision Agriculture and Environmental Technologies, Niagara College Research and Innovation.