Next steps ideas

[abiolaTresor]

Description

This issue is listing next steps that we may take to consider more complexity in Nevergrad optimization pipeline inputs.

Idea boxes

Short term (for eeia project)

1. Cartography: consists in generating a cartography for Benin showing how well rice culture performs depending the geographical location all over the country.
2. Explicative analysis: consists in deep analysis to find out which variables are explaining the most culture performances.
3. d-dimensional Optimization: consists in fixing all the parameters except one and give the optimal irrigation planning for that variable parameter. Then, switch to a space of two variables, etc..
4. Time Series Modelling: consists in modelling the time series represented by the Leaf Area Index (LAI) as a function of the SM (humidity index).

Mid/Long Term

1. Sowing strategy as input: consists in not assuming that the crops are already sowed before the optimization task. For example, let's consider that we have cereals $C$ whose sowing season is summer, let's say $4$ months. In our current optimization problem, we assume that all the cereals are cultivated at the beginning of the 4 months. But in practice, it may be interesting not to sow everything at the beginning but rather have a sowing strategy over the 4 months. Thus, we may try to see if we can inject sowing strategy information in the Nevergrad optimization approach here.

2. Stochastic resources as inputs: consists in considering that some of the parameters impacting the optimized cultures are not deterministic at the beginning and are being available stochastically throughout the culture period. For example, we may suppose that there is no predefined quantity of water at the beginning of the cultures but a quantity $X$ of water is available (through rainfall) everyday with probability $p$.

[abiolaTresor]

@akouete-kpakpo @Ethel2003

[akouete-kpakpo]

thanks for summarizing @abiolaTresor