Content uploaded by Julius Schöning
Author content
All content in this area was uploaded by Julius Schöning on Dec 03, 2021
Content may be subject to copyright.
©2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing
this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this
work in other works. The final publication is available at DOI 10.1109/GHTC53159.2021.9612460
AI-Based Crop Rotation
for Sustainable Agriculture Worldwide
Julius Sch¨
oning
Faculty of Engineering and Computer Science
Osnabr¨
uck University of Applied Sciences
Osnabr¨
uck, Germany
Email: j.schoening@hs-osnabrueck.de
Mats L. Richter
Institute of Cognitive Science
Osnabr¨
uck University
Osnabr¨
uck, Germany
Email: matrichter@uni-osnabrueck.de
Abstract—Artificial intelligence (AI) and sustainability. Two
words not commonly used in the context of crop rotation man-
agement. However, simple AI-based expert systems supporting
farms’ decision-making for optimizing crop rotation might be
the key to solving most of the UN sustainable development goals
worldwide. The essence of AI-based crop rotation and farm
management is that it works with nature—not against it! Thus,
the AI-based expert system needs to solve the multidimensional
optimization tasks of maximizing the diversity of crops that
match the local soil, local climate condition, the needs of the
livestock, and any available machinery. Next to the option task,
the expert system had to argue its decision-making basis since
the goal is to get the most sustainable farming with the most
profitable crop yield, not to get the largest yield. The expert
systems’ user interface (UI) should broaden thinking habits,
opting for new sustainable farming perspectives. This paper will
introduce both the technical architecture and the user-in-the-
loop (UIL) principle of such an expert system. Since our system
is still in the conceptional phase, we argue design decisions and
address open research questions needed to implement such an
expert system.
Index Terms—sustainable agriculture, AI-based farming, ex-
pert system, crop rotation, user interface
I. INTRODUCTION
How can farming cope with changing climate conditions
and extreme weather events? One quite old answer might be
even more intelligent crop rotation. The idea of crop rotation
is old, and even our ancestors recognize that changing the
kind of crops grown on the same field improves the yield and
plants’ health. The positive impacts of crop ration are also
scientifically proven [1], [2].
However, by the pressure to maximize profits, farmers are
often trying to maximize the amount of yield per hectare and
deviate from crop rotation growing similar crops one season
after another on the same fields. In this worldwide trend, the
crop rotation periods drastically shorten from seven-year to
even growing crops in monoculture [2]. Putative from crop
rotation will increase external inputs like chemical fertilizer,
organic fertilizer, pesticides, and water. Next to the reduction
of external inputs, crop rotation protects the land from erosion
by rain and wind, since longer-term rotations, including pas-
tures or grass-leys which absorb heavy rain events better as
row crops like corn [3], [4].
The critical question is how agriculture provides food secu-
rity and availability for the next decades and relieves farmers
from the market forces producing higher yields year by year.
In our opinion, an open-source AI-based crop rotation expert
system is the solution, ensuring sustainable, profitable, and
secure food production. This expert system provides support
for the farmers in the decision-making on her/his farming
strategies because the system provides short-, mid-, and long-
term plans dynamically based on the implications of today’s
farming actions. Thus, the farmers can assess impacts of, e.g.,
a yield amount maximization in the current season to the
profits in the next three years depending on all factors like
the cost of chemical fertilizer and pesticides.
For introducing the first concepts, architecture, UI, and
research question needed to be solved for the implementation
of such an open-source AI-based crop rotation expert system,
this paper first discusses the considered variables of the
multidimensional optimization task crop rotation in Section
II. Like any machine learning (ML) algorithm, AI is based
on data; in Section III we discuss how data can be entered
in the expert system in consideration of farmers’ experiences
and knowledge and without the use of comprehensive and
expensive sensor networks. Through the UIL principle, farmers
and AI go hand in hand. How UIL can give expert advice
and why UIL has not already been used in the agriculture
domain is evaluated in Section IV. The current technical
architecture and first UI mockups for specific tasks are also
sketched in this section. By outlining this possible architecture
and UI, open research questions ranging from the domain of
soil science via botany to the visualization and interaction
with AI algorithms. Highlighting the interdependency of an
open-source AI-based crop rotation expert system and the UN
sustainable development goals, in Section V, we conclude this
paper.
II. CROP ROTATION ,AMULTIDIMENSIONAL
OPTIMIZATION TASK
Why is crop rotation a difficult task? In theory, crop rotation
should maximize the diversity of crops in a three to seven-
year rotation, ensuring a balanced nutrient content of the soil
[5]. The starting point of any crop rotation is the knowledge
that crops are nitrogen hungry, phosphorous hungry, which are
©2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing
this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this
work in other works. The final publication is available at DOI 10.1109/GHTC53159.2021.9612460
cash crops
weather
climate
market demands
...
synthetic and
organic
amendments
...
livestock
tillage
soil
cover
crops
watering
...
electronic
...
apps
...
machanic
mobile
connections
machinery
(a) frame of attention (green line) usually considered for optimizing crop
rotation management
cash crops
weather
climate
market demands
...
synthetic and
organic
amendments
...
livestock
tillage
soil
cover
crops
watering
...
electronic
...
apps
...
machanic
mobile
connections
machinery
(b) ideal frame of attention (green line) for optimizing crop rotation man-
agement
Fig. 1. Crop rotation—the big picture of a multidimensional optimization task, (a) dimensions inside the frame of attention currently considered for crop
rotation management by little computer support, (b) dimensions that might be understood and approved by farmers with AI-based expert systems.
potassium hungry, and nitrogen-fixing. Up to this level, crop
rotation, has only one dimension, the dimension of the cash
crop, marked as the yellow circle in Fig. 1.
The next dimension is the soil. The optimization task
becomes trickier with the soil since the crops like corn,
wheat, grass, clover, carrots, oat, and triticale must match the
local soil conditions. The market demands as the dimension
strongly coupled with the profitability of the cash crops are
the dimension needed for efficient crops rotation. The farmers
also must consider the needs of the livestock available, the
weather, the kind of cover crops, the available machinery, law
regulations, subsidies, tillage, and many more dimensions.
The frame of attention a farmer considers by manual, i.e.
without or with a little computer support, does not cover
all dimensions possible, exemplary illustrated in Fig. 1 (a).
By applying an AI-based crop rotation management system,
ideally all these dimensions and their interdependencies are
considered.
In the first development stage of the open-source AI-based
crop rotation system, the dimensions: cash crops, market
demands, soil, watering, cover crops, tillage, weather, climate,
livestock, and synthetic amendments and their interdependen-
cies will be considered. Significantly, the four dimensions:
soil, watering, cover crops, and tillage are crucial in ensuring
climate-resistant farming. The condition of the soil is one of
the key values for sustainable agricultural production [1], [6].
In the second stage of development, the dimensions: ma-
chinery and energy consumption should be considered, but
even for the first stage, there are non-existing open-source
and free expert systems available. On the contrary, most data
platforms for farming are firmly in the hands of vendors for
agriculture machinery [7]. Consequently, solving the multidi-
mensional optimization problem of crop rotation might not be
available for small family-run farms.
III. FARMERS’ EXPERIENCES AND KNOW LEDGE
In everyday life, more and more decisions are made on
screens [8], even in agriculture, farmers relied on information,
e.g., from online weather services. Nevertheless, in contrast
informatic
climatology
soil
science
data
science
sensor
science
chemistry
electronics
zoology botany
AI
user-in-the-loop
farmers’ experiences and knowledge
Fig. 2. AI-based expert systems need to incorporate interdisciplinary “bookish
knowledge” combined with farmers’ experiences for proposing explainable
farming options which earn farmers’ trust.
©2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing
this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this
work in other works. The final publication is available at DOI 10.1109/GHTC53159.2021.9612460
to other domains, the most important information is gained
by framers’ observations, experiences, and knowledge about
her/his framing grounds. As illustrated in Fig. 2, the interdisci-
plinary “bookish knowledge” of the fields botany, zoology, soil
science, climatology, chemistry, data science, sensor engi-
neering, and informatics can be summarized easily by AI
and ML algorithms into applicable knowledge bases, and
ontologies [9]. The most challenging part, how to combine this
“bookish knowledge” with farmers’ practical and empirical
knowledge. Like in Fig. 2 the flower stems, the principle
interconnect “bookish knowledge” with farmers’ experiences
and knowledge.
Absorbing and process the farmers’ experiences and knowl-
edge about her/his fields and the current status of crop growth
can be measured by comprehensive sensor networks [10]
and be the use of guided questionnaires, like it is common
for evaluation of soil structure [11], [12]. The way how
smartphone apps like Feldgef¨
ugeansprache [12] interactively
support the farmers is one way of ensuring that even farms
worldwide will be capable of using a crop rotation expert
system.
Digital nudging—or why an AI-based crop rotation expert
system should be vendor-independent. Unfortunately, in more
and more software products, users are often engaged in making
buy-decision quite fast and in an almost automatic manner [8],
[13]. These kinds of paternalism must be avoided. One way
of doing so is to have an independent expert system, where
any company might profit by, e.g., selling fertilizer, pesticides,
sensor sets, and machinery.
IV. USE R-IN-THE -LOO P OF SU STAINAB LE FARMIN G
With their very fast comprehension, experience, and ana-
lytical as well as reliable problem-solving techniques, humans
could help AI achieve a broad breakthrough in sustainable
farming. In this case, interaction with users requires a simple,
consistent, and intuitive visualization technique so that even
inexperienced users can understand and comprehend the algo-
rithms and contribute their experiences to the algorithms in an
appropriate, scalable dialogue.
The UIL principle is indispensable for a functioning human-
machine cooperation. Thereby AI algorithms are the working
horse of the expert system, and they are processing data
from needed dimensions like weather, law regulations, amount,
and qualification of employees as shown as in Fig. 3. If
needed information is missing, the algorithms themselves must
recognize it, and the algorithm to ask the farmer for assistance
for filling this missing information, for instance, that the
farmer needs to input the crop growth for a specific field.
Consequently, the crop rotation expert systems are capable
to generate short-, mid- as well as long-term crop rotation
plans. Next to recommended farming action, all three plans
contain estimations of yield, water, fertilizer, pesticides, and
profit. By having plans of three different durations, impacts on
specific farming actions, like switching to a no-tillage strategy
AI-based
crop rotation planner
short-term plan (12month)
mid-term plan (36month)
long-term plan (180month)
domain
knowledge
weather law employee
...
user-in-
the-loop
Fig. 3. High-level illustration of the user-in-the-loop principle for crop
rotation. The expert systems need to provide a short, mid, and long-term
plan for crop management, at last, for dynamically providing the possible
implications of today’s farming actions.
[14], and its ecological, economic and technological benefits
become apparent.
For the worldwide applicability of the AI-based crop ro-
tation expert system, the UIL principle is essential, ensuring
that region-specific application of crop rotation concepts will
be considered. In addition, the UIL principle avoids that data
collected in another region of the world and based on other
farm sizes are erroneously used for giving expert advice. Sim-
plicity, i.e., providing AI-based systems without needing, e.g.,
mainframes, is quite important and challenging. Considering
rural small-scale farmers in developing nations, where the crop
rotation expert system might help achieve food security, it is
still changing to access AI cloud services. Therefore, instead
of developing a complex compute-intensive AI algorithm, the
architecture design had to target lightweight artificial neural
networks [15] that can be executed on smartphones and other
devices.
A. Give Expert Advised
For full acceptance, the users and the AI of software
need to work hand in hand, i.e., both must speak the same
language [16]. Only in this way can human-machine cooper-
ation succeed and the AI-based crop rotation expert system
can integrate farmers visual impressions about soil, crop,
and weed conditions into the computation of the rotation
plan. Giving experts advice, appropriate forms of interaction
must be developed, implemented, and tested. The working
environment has to be considered since, for instance, touch
screens are difficult to operate when driving in the field.
©2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing
this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this
work in other works. The final publication is available at DOI 10.1109/GHTC53159.2021.9612460
algorithms user-in-the-loop
algorithms decoder
(encode current status of
algorithms and its data)
knowledge encoder
(translate user knowledge
for algorithms and
current data state)
interaction interface
(visualization and
interaction)
domain
knowledge
input data
AI algorithm 1
AI algorithm 2
ML algorithm 1
ML algorithm 2
ML algorithm . . .
AI algorithm . . .
output data
Fig. 4. The user-in-the-loop principle, simple idea but hard to realize to the fact that AI and ML algorithms solve problems in high dimensional space which
must be visualized in such a way the expert user can interact with it.
Therefore, human-machine cooperation must take the context
of use into account and allow interaction via auditory, visual,
and haptic elements. Here, interaction methods such as voice
or gaze control could play an important role.
A first conceptual sketch of how the technical implemen-
tation of the UIL process in conjunction with AI algorithms
might look like is illustrated in Fig. 4. On the left side, the
classical software part is shown. This software part executes
hard-coded, trained processes in modern machines without
feedback from the user. Integrating the expert and experien-
tial knowledge of the users into the software, an algorithm
decoder, and a knowledge encoder are required.
The algorithm decoder acts as a translation aid so that
users with their very fast comprehension and analytical as
well as reliable problem-solving techniques can understand
the algorithm, including the data available at the current time.
The result of the translation process is visual and auditory
representations of the algorithms that users can see and com-
prehend.
If the users understand the current system state, they can
communicate their experiences to the algorithms in an ap-
propriate human-machine dialogue. The knowledge encoder
handles this knowledge’s feedback into a mathematically
understandable representation back to the algorithms. Imple-
menting the algorithm decoder and the knowledge encoder
in conjunction with AI algorithms is still a young field of
research.
B. Why hasn’t the User Already Put in the Loop?
Putting the user in the loop sounds quite simple, but
especially AI algorithms, as high-dimensional problem solvers,
are still opaque boxes, which are neither self-evident nor
human interpretable. Enabling the users to work closely with
algorithms understandable simplified visualizations of interim
results, as shown by Lundberg and Lee [17], are needed. These
visualizations project the data into low visualizable subspaces
and leverage the UIL principle within the agricultural domain.
Nevertheless, mapping data from all dimensions illustrated
in Fig. 1 to low visualizable subspaces is critical for expert
systems because these drastic simplifications in the model go
along with risking sub-optimal and miss-interpretations of the
results [18]–[20]. Thus, the farmer as the expert will get a
biased representation of the data and might question the advice
given.
In agriculture, many farming actions are based on the farm-
ers’ senses. For instance, for evaluation of the soil structure,
farmers’ visual, olfactory, tactile senses are needed. Thus,
the UI must provide an input mechanism ensuring a usable,
easy input of data. A possible setup UI mockup for the task
of evaluating the soil structure is illustrated in Fig. 5. For
analyzing the soil structure, the farmer needs to dig out a
spade-sized block down to a depth of approx. 30cm. In the
next step, by the use of her/his hands, the block start is broken
up, and the details about the soil fragments are documented.
For the digitization of this task, the UI must be entirely
designed for hands-free operation. The recent developments in
the visualization of intermediate results within AI algorithms,
as well as the emerging usability of hands-free UIs, make the
implementation of the UIL principle in the next few years
possible.
V. CONCLUSION
With the emergence of usable hands-free UIs an easy and
comfortable opportunity is given to incorporate farmers’ ex-
periences, knowledge, and observation into an expert system,
without the need for expensive sensor sets. Solving the mul-
tidimensional optimization task of crop rotation, the “bookish
©2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing
this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this
work in other works. The final publication is available at DOI 10.1109/GHTC53159.2021.9612460
aroma output device
for olfactory references
(optional)
voice assistant
on farmers’ pocketed
smartphone (use of
headphone optional)
Fig. 5. UI mockup for the task of evaluating the soil structure. The UI uses
a farmers’ smartphone, where a voice assistant is used for guided input of
the data. Thus, the farmer has her/his hands for performing the soil structure
analysis. If needed, olfactory references can be created by an aroma output
device, and visual references might be described in an auditory manner.
knowledge” is combined with farmers’ experiences using the
UIL principle. The UIL principle is proposing explainable
farming actions which earn farmers’ trust due to its explain-
ability. Therefore, the introduced first concepts of an open-
source AI-based crop rotation expert system might convince
farmers to continuously revising her/his farm management.
Since our open-source AI-based crop rotation expert system is
not implemented yet, we cannot provide results in numbers,
but by arguing our concepts and architectures in this early
conceptional phase, we are addressing open research questions
needed to implement such an expert system in the upcoming
years.
Note the term open-source expert system, since current yield
and farming data platforms are mostly vendor depended [7].
Therefore, the open-source expert system will lead to vendor-
independent short-, mid-, and long-term corp rotation plans
focusing on food security and soil quality of farmers’ fields
for the next decades.
Farming activities have mutual interaction with many UN
sustainable development goals. For naming the most important
ones only: zero hunger, clean water, responsible consump-
tion and production, and climate actions. By concluding this
paper, we are optimistic that an open-source AI-based crop
rotation expert system will establish sustainable agriculture
everywhere, since viable agriculture works with nature—not
against it!
REFERENCES
[1] S. Ouda, A. E.-H. Zohry, and T. Noreldin, Crop Rotation. Springer
International Publishing, 2018 . doi: 10.1007/978-3-030-05351-2
[2] A. J. Bennett, G. D. Bending, D. Chandler, S. Hilton, and P. Mills,
“Meeting the demand for crop production: the challenge of yield decline
in crops grown in short rotations,” Biological Reviews, vol. 87, no. 1,
2011 . doi: 10.1111/j.1469-185x.2011.00184.x
[3] S. P. Seibert and K. Auerswald, Hochwasserminderung im l¨
andlichen
Raum. Springer Berlin Heidelberg, 2020 . doi: 10.1007/978-3-662-
61033-6
[4] P. Fiener and K. Auerswald, “Influence of scale and land use pattern
on the efficacy of grassed waterways to control runoff,” Ecological
Engineering, vol. 27, no. 3, 2006 . doi: 10.1016/j.ecoleng.2006.02.005
[5] Earth Observing System. (2021, 6) Crop rotation: A way
to boost your yields. [Online]. Available: https://eos.com/blog/
crop-rotation- a-way- to-boost- your-yields/
[6] A. M. Bondarenko, L. S. Kachanova, A. V. Baryshnikov, and S. A.
Novikov, “Technologies for the production and application of organic
fertilizers in agriculture,” in The Challenge of Sustainability in Agricul-
tural Systems. Springer International Publishing, 2021, pp. 897–906 .
doi: 10.1007/978-3-030-72110-7 98
[7] Bundesministerium f¨
ur Wirtschaft und Energie. (2021, 6) Projekt: Agri-
gaia. [Online]. Available: https://www.bmwi.de/Redaktion/DE/Artikel/
Digitale-Welt/GAIA-X-Use-Cases/agri- gaia.html
[8] T. Mirsch, C. Lehrer, and R. Jung, “Digital nudging: Altering user behav-
ior in digital environments,” in 13. Internationalen Tagung Wirtschaftsin-
formatik (WI), 2017, pp. 634–648.
[9] D. O'Leary, “Using AI in knowledge management: knowledge bases and
ontologies,” IEEE Intelligent Systems, vol. 13, no. 3, pp. 34–39, 1998 .
doi: 10.1109/5254.683180
[10] T. Rajasekaran and S. Anandamurugan, “Challenges and applications
of wireless sensor networks in smart farming—a survey,” in Advances
in Intelligent Systems and Computing. Springer Singapore, 2018, pp.
353–361 . doi: 10.1007/978-981-13-1882-5 30
[11] The Royal Society. (2021, 6) Soil structure and its
benefits. [Online]. Available: https://royalsociety.org/topics-policy/
projects/soil-structure- and-its- benefits/
[12] Landwirtschaftsverlag M ¨
unster. (2021, 6) Apps f¨
ur die
feldgef¨
ugeansprache. [Online]. Available: https://www.dlg-mitteilungen.
de/mediathek/apps/
[13] M. Weinmann, C. Schneider, and J. vom Brocke, “Digital nudging,”
Business & Information Systems Engineering, vol. 58, no. 6, pp. 433–
436, 2016 . doi: 10.1007/s12599-016-0453-1
[14] R. Lal, D. Reicosky, and J. Hanson, “Evolution of the plow over 10,000
years and the rationale for no-till farming,” Soil and Tillage Research,
vol. 93, no. 1, pp. 1–12, 2007 . doi: 10.1016/j.still.2006.11.004
[15] M. L. Richter, J. Sch¨
oning, A. Wiedenroth, and U. Krumnack, “Should
you go deeper? optimizing convolutional neural network architectures
without training,” Under Review.
[16] J. Schoning, X. Jiang, C. Menon, and G. Heidemann, “Content-aware
3d reconstruction with gaze data,” in 2017 3rd IEEE International
Conference on Cybernetics (CYBCONF). IEEE, 2017 . doi: 10.1109/cy-
bconf.2017.7985769
[17] S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting
model predictions,” in International Conference on Neural Information
Processing Systems, ser. NIPS’17. Red Hook, NY, USA: Curran
Associates Inc., 2017, p. 4768–4777.
[18] I. E. Kumar, S. Venkatasubramanian, C. Scheidegger, and S. Friedler,
“Problems with shapley-value-based explanations as feature importance
measures,” in International Conference on Machine Learning (ICML),
ser. Proceedings of Machine Learning Research, vol. 119. PMLR, 2020.
[19] J. Sch¨
oning and G. Heidemann, “Visual video analytics for interactive
video content analysis,” in Advances in Intelligent Systems and Com-
puting. Springer International Publishing, 2018, pp. 346–360 . doi:
10.1007/978-3-030-03402-3 23
[20] P. Tanisaro, J. Sch¨
oning, K. Kurzhals, G. Heidemann, and D. Weiskopf,
“Visual analytics for video applications,” it - Information Technology,
vol. 57, no. 1, pp. 30–36, 2015 . doi: 10.1515/itit-2014-1072
