The Indus Village model

Modelling population, agriculture and climate change in the Indus Civilisation

Andreas Angourakis, Jennifer Bates, Jean-Phillipe Baudouin,
Alena Giesche, Joanna Walker, M. Cemre Ustunkaya, Nathan Wright,
Ravindra N. Singh and Cameron A. Petrie

available at https://andros-spica.github.io/RDMed-Angourakis-et-al-2021/
https://andros-spica.github.io/RDMed-Angourakis-et-al-2021/index.html?print-pdf (printable version)

1. The context: The Indus Civilisation

Avantiputra7, CC BY-SA 3.0, via Wikimedia Commons

One of the great ‘Old World’ Bronze Age civilisations

First cities in South Asia ('Mature Harappan')

Five known major urban centres

... and a multitude of smaller settlements

Mohenjo-daro, Sindh, Pakistan. "Great bath" and "granary". (Source)

Petrie & Bates 2017, J. J World Prehist

multi-cropping

Agriculture

Main crops:
- barley/wheat (winter)
- millets/rice (summer)
Pulses
Other

Bates, Petrie & Singh 2018, Archaeol Anthropol Sci

Other food production

Animal husbandry:
zebu, water buffalo, sheep, goat, pigs

Herding and specialised animal husbandry
( Chakraborty et al. 2018 , Lightfoot et al. 2020 )

Fishing, hunting, gathering

Bronze Age world system

T. C. Wilkinson 2014, TYING THE THREADS OF EURASIA, Fig. 7.4 | R. W. Law 2011, Inter-Regional Interaction and Urbanism in the Ancient Indus Valley, Fig. 13.6

Petrie et al. 2017, Current Anthropology

Urbanisation and
de-urbanisation

Green & Petrie 2018, Journal of Field Archaeology

Giesche et al. 2019, Clim. Past

summer rain, winter rain

J.-P. Baudouin (PhD dissertation)
summer-rain

Climate change

Giesche et al. 2019, Clim. Past

Urban phase
(Mature Harappan, c. 4.5 to 4.3 ka BP)
→ stronger winter precipitation

End of urban phase
(Mature to Late Harappan, >c. 4.1 ka BP)
→ decrease in both the summer and winter precipitation

2. Overview

Image source: Minecraft Wiki (© Mojang Studios), via Gamepedia

End-goals

Explore human adaptation to the variability
in landscapes of NW India during the Mature Harappan

Expose the sustainability of food production regimes,
mainly in terms of cropping strategies,
in front of abrupt climate change

Is diversity...

Adaptative? (How? When?)

detrimental to surplus generation and urbanisation?
(Good for villages is bad for cities?)

Rationale

rural settlement(s)
local scale (max. 25 km², 2500 units of 1 ha)
daily iterations
food production/consumption
Implementation in NetLogo, documentation in pseudocode diagrams and R markdown
Explore parameters (sensitivity analysis)
Explore scenarios
(procedural generation, alternative designs)

Before we started to build a model, we laid out a series of criteria to get us started on the right foot. | To properly answer our questions, we needed to represent rural, food-producing settlements, that nonetheless are capable of becoming larger settlements. | Even with potential growth, we are aiming at a strictly local scale, setting a maximum of twenty-five kilometres squared or two thousand five hundred units of one hectare. | The model iteration is set to a day, though some processes will only happen once or a few times a year. | Our interest is, for now, to only represent explicitly the essential elements of the Indus food system, which admittedly does not make justice to the Indus social complexity. | The model is being implemented in NetLogo in a modular and progressive manner, while each part is also thoroughly documented with pseudo-code diagrams and R markdown demonstrations. | To properly understand the model at each stage of complexity, we are performing extensive sensitivity analyses. | To inform and test our hypotheses, we will perform experiments design to represent scenarios, particularly by varying procedural generation parameters and using alternative designs.

Entities

	Household: propinquity and co-residence, rather than kinship
	Group: set of households, united by kinship or alliance; one or more may form a 'settlement'
	Land units: vary in elevation, soil properties, soil and surface water, cover and land use

To define what we believe are essential elements of a local-scale food system, our approach was to re-imagine an integral, yet blurry picture of what village life in the Indus would look like, based on known archaeological evidence, but also on minimum assumptions informed by parallels in space and time. In a broad sense, these elements include: | Climate; more specifically the weather conditions related to solar radiation, temperature and precipitation. | Land; which, as I mention, includes soil, water and cover. | Food production; mainly crops, which translate the condition of land and climate into foodstuff. | A human population; organised in households and groups and inhabiting a settlement. | The demography rulling the dynamics of this population, | The nutritional budget that regulates its health, | The diet that feeds into it, | and the food production economy that describes how production, distribution and storage of foodstuffs happens. | A exchange system that connects agents to other agents, and to agents in settlements outside the local economy. | And finally, bonding all together, the decision-making process that households would go through to chose and revise their food production strategies, particularly in terms of the triplet use-conditions-investment.

Modular design

Weather model
Soil Water Balance model
Land model
Crop model
Household Demography model
Storage model
Nutrition model
Exchange model
Household Position model

Why have a weather and land model at all?
Why not use contemporary or proxy-based data?

Issues:

precision
anachronism
data-driven simulations

However, see Contreras, Guiot, Suarez and Kirman (2018)

We identified the need for simulating weather and land variables, rather than using estimated data or real contemporary data. Why? | Well, we considered that there are a few deep issues that cannot be solved within the scope of our project: | Although there are a incrising number of paleoclimate and paleoenvironmental proxies available, the precise and absolute measure of weather and land variables in specific locations 4000 years ago is probably unreachable. | On the other hand, comtemporary data is, in someways, always underlying model design and parameter calibration. However, the use of contemporary datasets as estimations of conditions so far in the past is obviously problematic---we are talking here about climate change. | Moreover, the use of ANY dataset limited in scope as an input to a simulation model is a potential pitfall, given the perils of overfitting and circularity. | There are however promising approaches available that could help ease those problems.

Angourakis A, Bates J, Baudouin J-P, Giesche A, Ustunkaya M C, Wright N, Singh R N and Petrie C A. 2020.
How to ‘downsize’ a complex society: an agent-based modelling approach to assess the resilience of Indus Civilisation settlements to past climate change.
Environmental Research Letters, 15:115004.
DOI: 10.1088/1748-9326/abacf9

3. Weather model

Image source: blackreaper709 (© Mojang Studios), via reddit

Weather model | Target output

Angourakis et al. (in preparation, A), data source: NASA Power

Weather model | Submodels

annual sinusoidal curve
solar radiation, temperature

double logistic curve
precipitation (annual cumulative → absolute)

Angourakis et al. (in preparation, A)

Weather model | Dynamics

Angourakis et al. (in preparation, A)
files available at the Indus Village model repository

Weather model | Validation

Angourakis et al. (in preparation, A)

4. Soil water model

Image source: Marsh Davies (© Mojang Studios)

Gill et al. 2013, Urb. For. & Urb Green., Fig. 1

Adapted from
the ARID Soil Water model
Wallach et al. 2014,
Working with Dynamic Crop Models,
pp. 24, 138

One-Layer bucket soil model
R implementation published
Outcomes Reference Index for Drought (ARID)
Woli et al. 2012, Agronomy Journal, 104(2):287-300

Soil Water model | Dynamics

Angourakis et al. (in preparation, B),
files available at the Indus Village model repository

Precipitation and water stress

A point of comparison...

Wallach et al. 2006, p. 120, Fig. 3.4

South France (2007)

Hissar, Haryana, NW India (1995)

Input data (Hissar) obtained at: NASA POWER

5. Land model

Land model | Target output

J. Walker (PhD dissertation)

Land model | Landform

features → noise + smoothing → XY slopes → valley slope

Land model | Flows

flow direction and accumulation (+river accumulation)

Algorithms adapted from Huang & Lee 2015 , Jenson & Domingue 1988

Once elevation has been established, the Land model calculates the direction of flow for each land unit. This direction always points to the lowest of up to eight neighbous in the grid or towards the edges of the map. Land units that are the lowest among their neighbours, also called sinks, are filled up following the Huang & Lee algorithm, until there are only sinks present at the edges of the map. With flow direction, the model can then calculate a measure of flow accumulation also for each land unit, expressing how many land units flow directly or indirectly to it. On the left, you have a demonstration of this done in R. Modifying this calculation, an arbitrarily large flow accumulation (controlled as a parameter) is added to a land unit at the highest edge side and its flow direction is pointed towards the inside of the grid. With this added feature, we aimed at representing a passing stream or river without explicitly modelling its entire basin.

Land model | Soil depth and texture


depth	texture (% sand, silt, clay)	texture type (USDA)

Land model | Ecological communities


composition (% grass, brush, wood)	cover type

A similar logic is applied to generate the composition of ecological communities within each land unit, which is expressed as relative proportions of grass, brush and wood type vegetation. In this case, however, we assume a order of ecological succession where wood components take precedence over brush, and brush over grass. This configuration is to serve as the initial state of ecological communities and the climax of any further ecological succession, in case it changes due to perturbations. The immediate role of ecological communities is to inform a classification of cover types, which can be related how permeable to water is the land unit surface. Additionally, ecological communities form the base for estimating the subsistence value of those land units not used for cultivating crops.

Land model | Interface

Angourakis et al. (in preparation, B),
files available at the Indus Village model repository

The NetLogo implementation of the Land model is a fully-independent model that can generate terrains with a wide variety of characteristics, which can be explored in a controlled way through a set of meaninful parameters. Even though terrain generation can take less than a second, we added an import/export functionality that can be used to generate entire terrain libraries, regardless of when these will be used in simulations. We also made efforts in facilitating the visualisation of all generated aspects, particularly with multiple modes of patch colouring. At the bottom of the interface, we are able to monitor how a certain parameter configuration affects the calculation of soil depth (left), soil texture (middle) and ecological communities (right), depending on flow accumulation.

I1. Integrated Land Unit model

UPDATE: Land Unit model became Land Water model

To be consistent with the modular approach, we established our first modelling milestone as the Integrated Land Unit model, which combines the Weather, Soil Water and Land models into a working independent model. The challenge was the integration of the Land model with the Soil Water model, which in turn already integrated the Weather model. The Land model generates static terrain data while the Soil Water model describes a dynamic process within each land unit separately. Most inputs generated by the Land model are useble directly. However, there are several aspects that are not specified enough in either of these models, and thus required further design. Namely, the estimation of soil water capacities according to soil texture, how runoff exchange and inundation exchange are solved, and the impact of water on the ecological communities and cover types of a land unit.

Angourakis et al. (in preparation, B)
files available at the Indus Village model repository

The most important outcome of the Integrated Land Unit model is, of course, the ARID coefficient. Here you can see a simulation run illustrating how ARID reacts differently to precipitation and the influx of a passing river, depending on the different conditions of each land unit. Darker colour indicates lower water stress. With ancient agriculture in mind, this partial model already raises our attention towards the importance of a river inflow and a terrain allowing the formation of inundation areas. Note that, in this simulation, river inflow is kept constant, making it a more reliable water source than precipitation. Alternatively, this model can be easily extended to also represent seasonal streams. Feel free to pause this recording now to watch the entire video. ...

6. Crop model

Crop model

Angourakis et al. (in preparation, C), based on the SIMPLE crop model (Zhao et al. 2019)
files available at the Indus Village model repository

Expected relationship between ARID and crop yield

Angourakis et al. (in preparation, C)
files available at the Indus Village model repository

I2. Integrated Land Crop model

Angourakis et al. (in preparation, C)

Expected relationship between ARID and crop yield (II)

Angourakis et al. (in preparation, C)
files available at the Indus Village model repository

Conclusions

Modelling for the long haul
K.I.S.S. at your discretion
→Balance simplification and complexity
There are many models "out there" (replicability)
Importance of the cycle design-document-refactor
Share and document (as if there was no tomorrow!)

I would like to conclude this talk with a few takeaways from our experience developing the Indus Village model so far. | First, I insist on the importance of modelling for the long haul. Nowadays, most simulation modelling, and particularly agent-based modelling, is being done for short-term objectives, normally meaning they should begin and end within a few years time. Any academic endevour is inescapably framed by funding and deadlines fixed by institutions. Still, we should always keep in mind that there are certain questions (probably most of them) that are worth a lifetime of research. Modelling socio-environmental systems of the past is no exception. Therefore, I would argue for a certain attitude of individual resistance in this field, where we fullfill institutional and career requirements, but place the priority on contributing for long-term, collective and probably multigenerational challenges. | Secondly, I would like to advocate for following the KISS or "Keep It Simple, Stupid!" principle at your own discretion. Indeed, this principle is useful from a designer or engineer point of view; a simpler system IS generally less likely to break. However, in this field, we are not only creating artifical systems, but actually also trying to reproduce aspects of natural systems. These are complex, organically-evolved systems which are often unnecessarily complicated, filled with reduncies and hardly optimised for predefined purposes. As modellers, we should not try to include all possible elements of such systems in a single model. Even so, we should not ignore the complexity that we know to exist and should maintain a balance between simplicity of design and complexity of content. | As you may have noticed throughout this presentation, the Indus Village model relies greatly on models created and published by others. That has been an intentional strategy from the start. I believe that, given the limitations in time and funding for any given model, the combination of models offers a push for developing larger and better models, able to represent with higher fidelity the complexity of natural systems without becoming unintelligible. There are many models already out there, published in various formats. To find them, we just need to look! However, for model re-use to be possible, modellers in all fields must try to garantee the replicability of their work, which is no trivial task. | This brings me to the next point, which is the importance of following the cycle design-document-refactor. Remember, designing simulation models is also, and maybe above all, a craft involving a good share of programming. As so, it requires a great deal of trial and error, testing and not being afraid of taking steps back. Moreover, because our focus is not as much on performance as on intelligibility, we must force ourselves to constantly work on code documentation. These should range from extensive comments on code to high-level explanations in textual and graphical form. | Last, I call anyone involved in modelling to share and document their work, as if there was no tomorrow! Only through this kind of practice can we collectivelly achieve a truly model-based science capable of dealing with complex systems.

'TwoRains' project

ERC, 2015-2020

Acknowledgements

Thanks to the Land, Water and Settlement and TwoRains teams:

Aftab Alam, Alessandro Ceccarelli, Sagorika Chakraborty, Sudarshan Chakradhari, Arti Chowdhary, Yama Dixit, Charly French, Adam Green, Henry Green, Lily Green, David Hodell, Penny Jones, Carla Lancelotti, Emma Lightfoot, Frank Lynam, Sayantani Neogi, Hector Orengo, Arun Kumar Pandey, Danika Parikh, Vikas Pawar, Amit Ranjan, David Redhouse, Dheerendra Pratab Singh, & Akshyeta Suryanarayan.

Special thanks also to the Department of AIHC and Archaeology, BHU, the European Research Council (ERC), and the UK-India Education and Research Initiative (UKIERI) for support and funding, and to the Archaeological Survey of India for permission to carry out the work.

THANK YOU FOR YOUR ATTENTION!

available at https://andros-spica.github.io/RDMed-Angourakis-et-al-2021/
https://andros-spica.github.io/RDMed-Angourakis-et-al-2021/index.html?print-pdf (printable version)