Brewing a cassava crop model

^{Wordcloud of main concepts of this research approach}

All crop models are mathematical abstractions of the real-world interactions between crops and their environment. Because they differ in their purpose, they also differ in what components they include or do not include. For instance, modelers may choose to simulate (or not) nutrient or water limitations, depending on whether or not these factors constrain the system they are attempting to simulate.

Not only purpose but also our understanding of crop-environment interactions conditions what is or isn’t included in models. Such is the case for cassava.

Massive unrealized potential

Cassava is the second most important food crop in least developed countries and the fourth most important in developing countries, with total production of 218 million tons per year. In Asia, cassava production has nearly doubled in the last 10 years.¹

The “Rambo root,” as it is often called, is able to grow under dry and low soil fertility conditions.² Cassava is able to tolerate prolonged drought coupled with high temperature and low atmospheric humidity by adjusting itself to maximize water use efficiency. It is also capable of growing well at low rates of nutrient supply. Its hardiness makes it a key source of carbohydrate and income for millions of resource-limited smallholders around the world. In Southeast Asia alone, nearly 8 million farmers grow cassava, with 3 million of these in the greater Mekong Delta region (Thailand, Vietnam, Laos, and Cambodia).

Despite the crop’s importance, pests and diseases and suboptimal management imply more than a 75% difference between potential yields (i.e., attainable at experiment stations) and farmers’ yields.² Also, because of its hardiness, cassava is of particular importance for climate change adaptation when other crops such as maize or beans become unsuitable, as has been reported recently.

^{Cassava yield due to various inputs. Taken from CIAT. (2015)2}.

There is thus substantial potential to increase cassava crop production with better agronomic management and to expand production areas to new environments. Yet, the question remains: how to systematically produce recommendations on where and how to grow cassava at scale? Answering this question will eventually lead to more food availability and income for rural households around the world and specifically in Southeast Asia, which features the greatest diversity in uses of the crop (food, feed, starch for industry, alcoholic beverages, and even as an ingredient for noodles).

A crop model for improving cassava agronomy at scale

Five years ago, with the establishment of the Cassava Model Improvement Team (CMIT), we were convinced that with a process-based crop model we could help farmers, governments, and industry tap into existing knowledge to improve agronomic management at scale. And we still are!

In a joint effort, CIAT, the Thai Nguyen University for Agriculture and Forestry, the Soils and Fertilizers Research Institute (SFRI), and the Hung Loc Agricultural Research Center are set to improve the existing cassava simulation model within the Decision Support System for Agrotechnology Transfer (DSSAT^®). The project, started in 2016 and set to finish by the end of 2017, is funded by the Global Center for Food Systems Innovation (GCFSI) at Michigan State University.

At the time of framing the project, we learned two things that would define our pathway:

• First, no cassava crop model exists that ably simulates response to water and nutrient stress conditions.
• Second, the current model in DSSAT^®simulates cassava using principles similar to those for annual grain crops (e.g., wheat), which are not applicable to cassava, and thus needs a major re-engineering.

We therefore seek to make the best possible representation of the cassava cropping system in a model that will allow us to run simulations to predict the behavior of the crop under varied management assumptions (i.e., different planting dates, fertilizer application rates, and precipitation/irrigation regimes) and many different types of soil and climate (including those arising with climate change). Our project consists of four major activities:

• Documenting comprehensively the algorithms that describe the physiology of cassava, including its response to water and nutrient stress.
• Implementing such algorithms in the DSSAT^®computer code.
• Conducting new field trials to understand responses to nutrient and water stress (in Vietnam and in Colombia, respectively).
• Carrying out model calibration and testing.

Project team at Hung Loc Center discussing aspects of the field monitoring. From left to right: Nguyen Thi Thu Huong (research assistant), Nguyen Ba Tung (research assistant), Pham Thi Nhan (cassava team leader at Hung Loc), Julián Ramírez-Villegas (CIAT scientist, project co-PI).

Progress in model development

Decades of research in cassava physiology and breeding mean that strong knowledge of the biological behavior of the cassava crop exists. Our model considers several processes, including temperature-­driven development, the structure of the plant, planting dates, planting material, germination, and node development, among many other aspects. Mathematical equations in the model are used to represent these processes and their interactions.

These equations can be coded into a programming language in order to run simulations based on the inputs established for targeted experiments. In this case, the code is incorporated into the DSSAT^® source code (in Fortran), since it is a widely known software that contains crop simulation models and already has interfaces for reading the inputs required to run the model and also to export the outputs of the simulation.

The current coding stage started with 33 tasks in the backlog. Seventeen out of the 33 modifications planned are already in the model code and 16 are in the project backlog pending to code. Progress has reached 51% thus far. The major functional advances are the following:

• A spillover model without storage root initiation in a fixed time;
• The development of the plant as cohorts of nodal units, which approximates closely how the plant actually grows;
• Effects of air-to-canopy vapor pressure deficit (VPD) in transpiration and photosynthesis;
• Incorporation of a nitrogen distribution component in the model that simulates cassava ability to produce and grow well in infertile soils; and
• Improved water and temperature stress routines.

More improvements are on the way as we continue to mine existing data (see below) and finish our field activities in Vietnam and in Colombia. Based on these, two major releases of the model are expected:

• by January 2017, we will release a new version with all water stress components fully incorporated, and
• by June 2017, we will release a version with nutrient stress fully incorporated.

Data, the bigger the better

As we move from process documentation and model coding to calibration and testing, we see the need to have experimental data to comprehensively assess the model under as many real-world situations as possible. Unfortunately, despite the thousands of cassava trials conducted in different parts of the world by CGIAR, no trials are systematically documented for Asia.

An overview map of existing cassava trials around the globe, as queried from the AgTrials.org Repository on 16 October 2016. Note no trials reported for Asia.

Most importantly, perhaps, is the fact that we could recover reliable data for model development, calibration, and testing from only 21 trials (see our progress report here). This underscores even more the importance of our nutrient trials in Vietnam and of our water stress trial in Colombia.

Cassava team in Yen Bai Province in North Vietnam discussing aspects of the sampling method. From left to right: Julián Ramírez-Villegas (CIAT scientist, project co-PI), Mr. Ha Viet Long (lecturer, Thai Nguyen University), and Long’s student Tra.

Further: the cassava Khon Kaen University in Thailand, led by Dr. Poramate Banterng may be one of the very few groups doing basic research in cassava physiology. Our visit to them in October 2016 led to identifying many opportunities to feed new knowledge into the model.

Scientists from Thailand (Khon Kaen University and Chiang Mai University) and CIAT after an exciting day-long session about cassava crop modeling and agronomy, during a meeting held at Khon Kaen University on 17 October 2016.

So, stay tuned to catch up on our progress! A cassava simulation model is brewing and will be ready soon!

CIAT team

• Julián Ramírez-Villegas – co-principal investigator, crop modeling
• Tin M. Aye – co-principal investigator, agronomy
• Daniel Amariles – programmer
• Jonatan Soto – agronomist
• Patricia Moreno – agronomist
• Mayra Toro – agronomist
• James Cock – emeritus CIAT physiologist
• Myles J. Fisher – emeritus CIAT crop modeler and physiologist

Project partners

• Nguyen The Hung, Thai Nguyen University
• Nguyen Viet Hung, Thai Nguyen University
• Ha Viet Long, Thai Nguyen University
• Tran Minh Tien, Soils and Fertilizers Research Institute
• Da Trong Thang, Soils and Fertilizers Research Institute
• Nguyen Huu Hy, Hung Loc Agricultural Research Center
• Pham Thi Nhan, Hung Loc Agricultural Research Center

References:

¹https://faostat3.fao.org/home/E

²Centro Internacional de Agricultura Tropical (CIAT). 2015. Sustainable Cassava Production in Asia – for Multiple Uses and for Multiple Markets. Proceedings of the Ninth Regional Cassava Workshop, held in Nanning, Guangxi, China P.R. 27 Nov – 3 Dec 2011. 13 –  25.

Authors of this post:

Dr. Julian Ramirez-Villegas

co-Principal Investigator, crop modeling

I.A. Jonatan Soto Bermeo

Agronomist

I.S. Daniel Amariles

System Analist, crop modeling