We would like to cultivate a culture of discussion

...with our agriconBLOG


01.11.2018 - Peer Leithold (Send email to Peer Leithold)

Yield potential maps - Why they do not work

Yield potential maps - Why they do not work

POSITION - Yield maps have been known in agricultural practice for over 25 years. The measurement of field-specific yields and the subsequent creation of yield maps for a field has improved enormously during this time. These days their quality can therefore be rated as quite good. Now a question arises, what to do with yield maps? There are two possibilities:

  1. Deficiency analysis of the yield result of a year - good idea!
  2. Basis for drawing conclusions for the following year - hands off!


Why not?

From a series of yield maps, yield potential maps are to be calculated. These, in turn, are to be used in precision farming in order to plan management measures and the use of resources precisely. At first apperance this idea seems to make sense. Therefore, many farmers also dealt with this topic and invested, for example, in yield measurement systems in combine harvesters.

However, the wish was father to the thought! I can still remember the time around the turn of the millennium, when many users had access to multi-year yield maps for the first time. The sobering conclusion was: "The more yield maps we collect, the less we understand."

I beg your pardon? Does this mean that the more concrete the data is, the more clearly it becomes apparent that yield potential maps do not work?


Let's take a closer look! 

Prof. Simon Blackmore (Harper Adams University) investigated back in 2002 the possibility of predicting future yields using the structure of 6-year yield maps from Denmark. The aim was to produce yield potential maps in order to be able to derive future management measures. Prof. Blackmore came to five key statements:

  1. The differences in yield between the individual years can have the biggest influence on the yield.
  2. The spatial variability of yields within a field (yield pattern) in a single year is very pronounced.
  3. The yield patterns of several years cancel each other out.
  4. Yield trend maps cannot predict the yield of the following year.
  5. Therefore, instead of the yield the current demand has to be managed.


Yield level, yield pattern and predictability of future yield

Based on the exact scientific evidence of Prof. Blackmore, we repeated the analysis with yield maps from Germany. We examined 5- to 8-year-old yield maps of a total of seven fields from Mecklenburg-Western Pomerania, Saxony-Anhalt, Thuringia and Saxony. The exact question was whether future yields can be forecast reliably enough on the basis of historical yield data? Our results were amazingly clear!

However, when analysing the data, you must first distinguish between two effects:

Effect no. 1: Annual effect of yield formation or where is the average yield of the field?

à For each field you can calculate an average yield over a period of time. In our example data these average yields were between 54 and 89 dt/ha, depending on the field and the number of years considered. 

à But the exciting question is, how far can the individual years deviate from the long-term average yield of the field? In our sample the individual year conclusions on average over all fields by about 17 dt/ha (7 to 28 dt/ha) upwards and downwards. This means that the annual effect alone (good and bad years) leads to a yield variability of the average yields of fields in the order of about 34 dt/ha. Who dares to predict the average yield of a field in October or March in a quality of about 5 dt/ha? The one would be a made man in hail insurance or with brokers!


Figure 1: 8-year series of yield maps of a field

1998: 82 dt/ha 1999: 71 dt/ha 2000: 76 dt/ha 2001: 99 dt/ha
2002: 59 dt/ha 2003: 39 dt/ha 2004: 65 dt/ha 2005: 62 dt/ha


Figure 2: Mean value of the 8-year series of yield maps of a field

Midpoints 1998 - 2005

Note: Uniform legend for all years; 13 classes of 10 dt/ha each; 0-130 dt/ha; maximum blue, minimum yellow


Effect no. 2: Yield pattern of the single year or are there spatially distinct yield differences within the field and are they always the same?

The yield differences within a field and a year are more or less clearly marked. The degree of manifestation of this difference can be described mathematically by calculating the standard deviation. The standard deviation is a measure of the spread of the values of a characteristic around its average value (arithmetic mean). In our case it means that the standard deviation of the yield is the average distance of all measured individual yields from the average yield of the field. In our survey fields, the average standard deviation was around 12 dt/ha (5 to 28 dt/ha).

To put it simply, this means that on any given field, each individual yield of a sub-area is on average (!) 12 dt/ha up and down from the average value of the field. So if the average yield of a field was 80 dt/ha, then statistically speaking each yield point measured would be within a corridor of 69 to 92 dt/ha. Of course, many of these points are closer, but other yield points are even further away from the average value."

If you look at the yield patterns of the trend chart (average yield chart over time), they tend to cancel each other out. The more one averages, the less significant the yield patterns of the multi-year yield map are compared to the one-year yield map. Some yield maps of dry and wet years even cancel each other out completely. The question is, how exactly can these yield patterns, i.e. deviations from the average yield, be predicted? Assuming that an uncertainty corridor of around 10 dt/ha is accepted for planning purposes, the following result is obtained. Only three to a maximum of 20% of the areas can be predicted as stable high or stable low yield patterns. Looking at it the other way around, for 80 to 97% of the area no deviation of the individual area from the yield averageof the field can be predicted with sufficient certainty.


One question remains

Can we use multi-year yield maps to predict future yields for the individual year? We have tried to do so using this sample data. We have used the prediction quality (R²) as a benchmark. This is on average 0.15. This means that only 15% of the yield of a single year can be explained by the historical data. The best values were 41%, the worst values 0%. In other words, it is not possible to draw conclusions about next year's yield from the historical data.

Now, on the other hand, it can be argued that historical earnings data alone do not lead to a map of earnings potential. One should also include soil samples and weather data. I have been hearing this argument for a good 25 years now. But in exactly this time I have not found anyone who could show that it works anywhere. If you know, for example, that the soil type correlates positively or negatively with the yield or that nobody can predict a dry or a wet, or a cold or a warm year, then this approach fails in the field.

However, what you can find in relevant literature are scientific studies which, more or less come to the same conclusions, namely the non-predictability of future yields.



  1. From the consideration of multi-year yield maps, the future yield cannot be predicted with sufficient certainty for the derivation of operative plant management measures. The agronomic approach, i.e. planning the expenditure of resources on the basis of yield potential maps, cannot be supported (or only to a limited extent).
  2. You should concentrate on managing the current variability within one year. It is important to identify limiting factors during the growth process which limit the possible yield of a year (water supply, temperature, global radiation) and determine their effects (current N demand, infection pressure, occurrence of weeds etc.) on the use of inputs..
  3. Instead of the yield target, the current demand must be managed.
  4. This analysis does not question the general use of yield mapping systems. These are still an excellent tool for, among other things, the annual weak point analysis. However, the data generated with them cannot fulfil the function they were intended for.

It amazes me every time that the topic of yield potential maps finds its way into agricultural magazines and at agricultural conferences. In my opinion, its supporters have never had to put their claims to the test. It is and remains mere theory - without feasibility and accuracy. The fact that it is still being heard in the field is astonishing and sobering. Nor do I understand why yield potential maps are regularly mentioned in connection with plant sensors (not the YARA N-Sensor®!). Because I observe with concern that individual scientists, consultants and also sellers of these sensors always dump the task of creating alleged yield potential maps on the farmer. And if the calculation does not work out, it is of course not the fault of the plant sensor, but the farmer. The farmer simply did not produce a real yield potential map! You can make it that "easy" and continue with wishful thinking.


pdf | 273 KB

Write the first comment on this article

Your email address will not be published. Required fields are marked with *.

Back to listview

Please select your language

We have noticed that you are visiting the website with a different language. Please select your preferred language.