Overview

Use Case 1 of the COSA project targets one of the most foundational questions in agriculture: how can we accurately, dynamically and at scale evaluate the real potential of a piece of land and forecast the yield it can deliver? The use case combines classical agronomic land bonitation with modern Geographic Information Systems (GIS) and machine-learning models, and validates the resulting framework on real Romanian agricultural data.

Scope

The use case integrates several streams of work:

• Soil and land studies — pedological studies, land bonitation tables and historical cadastral data are processed and harmonised as input for the COSA framework.
• GIS-driven land evaluation — cadastral information, soil maps, topography and climate data are combined in a GIS environment to compute spatially aware bonitation scores at the level of individual fields.
• Machine-learning enhanced bonitation — adaptive models continuously analyse complex spatial and environmental data to evaluate land potential dynamically and to keep up with a changing climate.
• Crop yield prediction — supervised models predict crop yields based on agronomic and environmental indicators, supporting both farmers and policy-makers.
• Crop management optimisation — fertilisation and crop-management strategies are optimised for specific crops and regional contexts (e.g. triticale in the Lăpuș depression).

Methodology

The use case follows a layered approach consistent with the overall COSA architecture:

1. Data layer — pedological studies, land bonitation tables, cadastral and amenajament documents, GIS layers and weather time series are ingested and structured.
2. Modelling layer — feature-selection methods are used to identify the most relevant variables for soil moisture and precipitation prediction, and deep-learning artificial neural networks are then trained for environmental forecasting.
3. Decision layer — the resulting models feed into a GIS-driven, machine-learning-enhanced framework for adaptive land bonitation that produces up-to-date land quality scores and crop recommendations.

Background materials

Use Case 1 builds on a curated set of agronomic and methodological references, including: pedological studies (e.g. Tinosu commune), official land bonitation tables, methodological references on agronomic land evaluation, soil moisture and humidity algorithms, cadastral and amenajament documents, and prior doctoral research on land evaluation using GIS.

Live platform

The work of Use Case 1 is materialized in the Land Bonitation Platform, a publicly accessible GIS application developed by Bogdan Văduva in partnership with Vital.

Selected publications produced in this use case

• M. T. Sattari, A. Avram, H. Apaydin, O. Matei, «Evaluation of Feature Selection Methods in Estimation of Precipitation Based on Deep Learning Artificial Neural Networks,» Water Resources Management, Springer, 2023. (Q1)
• I. Cionca, A. D. Costin, T. Rusu, «Optimizing fertilization and crop management for triticale in the Lăpuș depression, Romania,» AgroLife Scientific Journal, 2024.
• C. Anton, A. Avram, O. Matei, L. Andreica, B. Văduva, «Advancements in Machine Learning Algorithms for Precision Crop Yield Prediction: A Comprehensive Review with Focus on European Union,» SOCO 2024.
• B. Văduva, O. Matei, A. Avram, L. Andreica, «Embedding GIS in crop field bonitation computation,» SOCO 2024.
• B. Văduva, A. Avram, O. Matei, L. Andreica, T. Rusu, «A GIS-Driven, Machine Learning-Enhanced Framework for Adaptive Land Bonitation,» Agriculture, MDPI, 2025. (Q1)
• I. Băărăian et al., «Trend-Enabled Recommender System with Diversity Enhancer for Crop Recommendation,» Agriculture, MDPI, 2025. (Q1)
• O. S. Mintas et al., «Sustainability of the Integrated Waste Management System: A Case Study of Bihor County, Romania,» Sustainability (Basel), MDPI, 2025.

Expected impact

• Precision crop matching — farmers can match crop varieties to the optimal soil conditions of each specific parcel.
• Dynamic adaptation — as climate, rainfall and soil conditions evolve, the machine-learning models adapt and continue to deliver accurate, future-proof land-quality scores.
• Sustainable land management — the data-driven approach helps prevent land degradation and over-farming, supporting productive and environmentally sustainable agricultural practices.

« Back to use cases

Publications related to UC1 — Crop Yield & Land Bonitation