Finding the sweet spot in the sugar cane fields
By Yasel Costa Salas, ZLC Professor.
The cane sugar industry is of vital importance to rural communities in many parts of the world, but it faces its own particular challenges. Sugar being a globally traded commodity, growers and processors have little price-setting power, and competition is fierce, not only between cane sugar producers but with the beet sugar alternative. As around 60% of total supply chain cost is incurred in upstream activities (growing, harvesting, and transport to processing mills), finding the right balance between manual and mechanised methods, and the optimal planning and scheduling of harvesting and logistics resources, is critical.
Together with colleagues from several Colombian universities, we have been developing an optimisation model for these activities. ‘Multi-objective stochastic scheduling of upstream operations in a sustainable sugarcane supply chain’ is a cumbersome title, but unpicking the terminology reveals some of the complexities and constraints involved.
Our modelling needed to reconcile three somewhat conflicting objectives, starting with minimising operational cost. Harvesting can involve manual cane cutting and stacking, using lifting machines to load vehicles to transport to mills. Alternatively, harvesting can be mechanised using a variety of machines, the choice of which may be constrained by the terrain. This requires far fewer workers, although cane choppers are still needed to break cane stalks to facilitate loading. Transport to the mills can involve various combinations of trucks or tractor units and trailers (traditionally, many fields used light railways, with steam engines powered by eco-friendly bagasse or dry mill waste, but this is now rare).
There are some important constraints on scheduling. Although the mills operate 24 hours a day, manual cutting is confined to daylight hours, in practice a single shift from 06.00 to 14.00 (although loading can carry on into the evening). Mechanised harvesting can work through, but it is a tough environment for machines, and so maintenance periods of up to 4 hours, again in daylight, need to be scheduled. The ability to stockpile cut cane is limited – if the cane isn’t processed within nine hours of cutting, quality and yields deteriorate significantly.
Minimising these operational costs – machinery costs, maintenance, transport, and especially wages – while fulfilling mill demand is an obvious goal but it is not the only one. A second, increasingly valued, objective, is to minimise negative environmental impacts, specifically CO2 emissions. This is less obvious than might be thought – of course, less use of machinery reduces fuel emissions, but manual harvesting requires the surplus vegetation to be burned off – sugar cane leaves are razor-sharp and 8impede manual cutting, while the fields can harbour rats and snakes. (Although not considered further in this model, burning also negates another environmentally beneficial supply chain, the supply of green harvest waste to ethanol-manufacturing plants).
The third objective is that of positive social impact – Job creation. Many communities in cane-growing areas are critically dependent on the harvest ‘campaign’ as one of the few sources of cash employment. The environmental and social goals explain ‘sustainable’ in the project title.
The term ‘stochastic’ also requires explanation. Many elements in the optimisation – the capacity of a given transport fleet, the area per hour that can be harvested mechanically or manually, and so on, are within limits fairly deterministic – but the key element, the yield in tonnes per hectare, is not. Yield is affected by soil and geology, and by the variety of cane planted, but particularly by any pests and by the weather throughout the growing season – the way these factors have interacted, and thus the harvestable yield per hectare, is essentially unknown until the harvest is underway. The higher the yield the more resource required to harvest; the lower the yield, the greater the area that needs to be harvested in a time period to keep the mill fed.
Our model therefore had to allow for this stochastic element. For a given parcel of land we analyse a baseline or expected yield scenario, based on crop variety, historical data, and so on. But we also look at high/low, or best/worst case, scenarios, with uncertainties of +/- 5%, 10% and 15%.
Given all this (and several other constraints) the task of the model is to derive a detailed schedule for the land parcel under consideration that finds an optimum balance between the three objectives, expressed in the answers to five questions:
- What is the size of the terrain to be harvested within this land parcel? (because it may not be optimum to harvest the whole parcel in one go).
- What type of harvesting method (mechanical, manual, or both) is best?
- In what quantity are these resources (machinery and workforce) required and to what schedule?
- When, and with what resources, is machine maintenance to be carried out?
- What quantities and combinations of transport equipment are required, and when, to move cut cane to the mills in timely fashion?
To tackle this, we created our stochastic multi-objective mixed integer linear model. The model does not, of course, produce a single ‘right’ answer – rather, it generates a set of efficient solutions, depending on the weighting applied to the three objectives (an efficient solution is one where you cannot further improve one element without significantly impairing others). This set of solutions forms a Pareto frontier, and there is a ‘Utopia’ point, which would be the optimum outcome across all three objectives. Unfortunately that may not be all that close to any of the efficient solutions, but Compromise Programming is a technique used to find the best fit.
The model has been tested and verified in real life cane fields in Peru. Mechanical harvesting methods perform better on cost, but require higher workforce skills and an investment that seeks economies of scale. Mechanical methods may also be superior in terms of air quality. Manual harvesting of course creates more jobs (although perhaps lower paid) and the model gives some insight into the economic penalty of this job creation and thus what sorts of incentive governments might wish to offer famers in areas of high rural unemployment. The model also reveals the degree of sensitivity of the ‘optimum’ solutions to the stochastic nature of crop yield.
For more information contact Yasel Costa at [email protected], or read the paper in the Journal of Cleaner Production, Vol 276, December 2020, at https://www.sciencedirect.com/science/article/abs/pii/S0959652620333503?via%3Dihub