SPATIAL architecture 

This workpackage involves the design of the architecture of the SPATIAL proof-of-concept prototype. The work involves specification of the scope and functionality for each of the system’s modules that serve as guidelines for the implementation of a Proof of Concept prototype. The prototype will follow a 3-tier architecture. The 1st-tier involves the specification of data stores for collecting the datasets to be used for training the ML-models for Soybean yields and futures contracts price moves forecasting. The 2nd-tier involves the 2 distinct modules implementing the forecasts. Considering that diverse models may need to be tested the architecture will foresee that both modules are black-boxes for which only API-based interfaces need to be specified (rather than specific implementation technologies) in terms of feeding them with input datasets and collecting outputs (forecasts) and chaining their outcomes into an overall pipeline. Finally, the 3rd-tier of the architecture involves the SPATIAL prototype front-end. The outcome of this workpackage will feed into WP2100 and WP2200 such that the interconnectivity requirements are respected and also to WP2300 which will take over the final integration and implementation of the proof-of-concept prototype.

Datasets Overview 

SPATIAL will integrate Earth Observation, geospatial, crop yield and financial datasets in order to achieve improved signals of soybean commodities prices rise or fall and related soybean futures contracts price move forecasts. Within this workpackage the relevant EO and non-EO datasets will be assessed in two ways: availability of archived data to build the AI models for crop yield and price move forecasts and the timeliness of the new data for meaningful prediction for the current year (growing season).

In the framework of SPATIAL a rich set of data from Copernicus will be exploited as input to the AI crop yield forecast model. These datasets with appropriate and targeted processing will form the basis to train the model (archived data) and be used as input to forecast the annual crop yield. Climate Change Service (C3S) data are expected to provide significant information in this respect. Regarding the financial datasets, that will be used as input to SPATIAL’s AI Soybean futures price move forecast model, we will use Yahoo Finance to source data related to a series of financial assets that affect Soybean futures prices and also additional sources such as the USDA the FRED etc.

Design and development of AI models for Soybean crop yield forecast 

As part of this workpackage, crop yield modeling and forecasting will be implemented through intensive use of Machine Learning algorithms that correlate the sole output variable to be estimated and eventually forecasted (annual crop yield) with timeseries datasets of several explanatory variables used as predictors.

Design and development of AI models for Soybeans futures contracts price move forecast

In SPATIAL the models which will be used for forecasting soybean futures contracts price moves will take into account multiple datasets after performing a thorough feature engineering for selecting the features with the highest impact. The initial set of data sources e.g. financial data variables and yield prediction data will be investigated using alternative ML models either in non-automated fashion or with the help of AutoML technology.

SPATIAL proof of concept modules integration and evaluation

The data storage layer along with the two forecasting modules of the SPATIAL proof-of-concept prototype as described in WP1100 and implemented in WP2100 and WP2200 will be functional and will serve as inputs to this workpackage. This workpackage will be responsible for developing the appropriate interfaces as well as the specified front-end in order to allow running of “end-to-end” forecast scenarios. An “end-to-end” forecast scenario corresponds to the user specifying the request for a forecast regarding the price move of a Soybean futures contract, which translates into sourcing the relevant datasets from the data storage and generating the yield prediction by the SPATIAL AI module for Soybeans crop yield forecasting, and then passing this yield prediction along with additional financial datasets into the SPATIAL AI module for Soybean futures contracts price move forecasting to return the final prediction to the end-user interface along with qualitative characteristics of the prediction (confidence level).