AgEngineering 🌾 | Computer Vision 👁️ | Spatial Data & GIS 🌍
I build deep-learning frameworks, digital twins, and self-supervised architectures for agronomic and environmental applications, combining GIS, remote sensing, and machine learning with rigorously validated, reproducible research code.
|
10,190 simulations of the Daisy model (University of Copenhagen), a soil-plant-atmosphere model, (10 soils, 17 experiments) mapping non-linear soil-atmosphere feedback loops between precision irrigation triggers, evaporation-control mulching, drainage, yield, and nitrate leaching. Contributions: a soil-independent irrigation trigger law (root-zone depletion collapses suction-based triggers from 10 soils onto single curves), a closed water-balance account of where evaporation saved by mulching actually goes, a regime map of when the dominant water-loss pathway switches from evaporation to drainage, a measurable soil hydraulic memory for the recoverable share of an evaporation saving, a closure-first audit of the water balance, and the SWAFL software built on them.
SWAFL Desktop 1.1.0: the drip irrigation design with the soil water of the days ahead (DSS), every day from the working date coloured by phase, the root-zone water against field capacity and the trigger, the next irrigation and the wilting point without irrigation, beside the Dashboard and the Properties. Click the picture to open the online version.
Field on the map (Tools, remote sensing): a real field demarcated on the satellite image (2.35 ha), its Sentinel-2 NDVI of 23 October 2026 cut to the outline, with the management zones of NDVI and their areas, the soil from SoilGrids, and the site analysis (water sources and the irrigation system suited to the field).
The graphical workbench for the Daisy model (menu Daisy): a guided path of nine steps from a question to a run, 17 prepared experiments with compact designs, the process map of the model, the soil column (Richards equation), response curves and calibration (sequential, CMA-ES, Bayesian), and the route of the results into the SWAFL decision layers. Daisy 7.1.14 is bundled for Windows. Software created from this research (SWAFL). SWAFL is built from the 10,190 Daisy simulations, about 470 numbered equations with a stated origin (literature, data-driven or engineered) and four manuscripts, all in this repository. It is an offline-first suite of 22 decision layers (irrigation design for surface, drip and sprinkler systems, irrigation control, mulch and evaporation, soil structure, field water balance, seasonal data, crops, feedback loops, deterministic phenotyping), a graphical workbench for the Daisy model and its bridge to the decision layers, tools for the real field (field sketch, Sentinel-2 indices, zone-by-zone variable-rate irrigation, field report) and a library of 48 crops. The weather is a choice (a benchmark record, seven bundled stations, the user's own file, online weather of any place or DMI open data), and a question written in plain words becomes a checked Daisy experiment. It computes and shows the trigger law, the mulch water account and the regime map from this work, and it can be used without installing anything. No field validation is claimed. Software files (version 1.1.0): Software/v1.1.0 folder of the repository
Run online (virtual version, nothing to install): SWAFL Desktop 1.1.0 in the browser |
|
Operationalizing GIS & ML Across Cropping Systems Maize fertility zoning, vineyard pest monitoring, and grape ripening prediction across three linked Portuguese field practicals. 
|
|
Geometry-Agnostic Contrastive Learning (GACL) A hypergraph-transformer architecture for cross-pathology, cross-condition contrastive transfer learning in agronomic imaging. 
|
|
Real-Time RGB Vegetation Indexing (UAV/Handheld) Real-time RGB proxy vegetation indexing and texture analysis pipeline for UAV and handheld crop imagery, with an integrated GIS workbench. 
|
|
A modular Python pipeline with a Tkinter desktop GUI for reference/crop evapotranspiration, soil-water balance, irrigation scheduling, and a 4-step crop-growth simulation (canopy, transpiration, biomass, yield) for five crops grown around Zaria, Nigeria. Renamed and extended from the Zaria Crop ET and Irrigation DSS pipeline above. 
Field-stage simulation with per-stage water accounting. |
|
GIS Monitoring of Vineyard Ripening (Brix) NDVI and logarithmic regression harvest-date forecasting (R-squared = 0.9784) for grape ripening at a 6-hectare UTAD vineyard, Quinta de Nossa Senhora de Lurdes, Vila Real, Portugal. 
|
|
Terrain characterisation, mating-disruption costing, and delta-trap monitoring for Lobesia botrana at a 42.97-hectare vineyard estate in the Douro Demarcated Region, Portugal. 
|
|
Digital Twin for Gully Biocontrol Bayesian-grounded digital twin evaluating Morning Glory (Ipomoea spp.) as a gully-erosion biocontrol measure, validated against real field-sensor data. 
|
|
Hybrid CNN-BiLSTM-Attention for Sediment Transport A from-scratch NumPy deep-learning framework for sediment-transport prediction in an agricultural gully system. 
|
|
Geospatial Data Analysis Course Package, vector/raster GIS, Sentinel-1/2 remote sensing, and classification (Jupyter notebooks + Google Earth Engine) 
|
|
Multitask Singularity Regression Kriging (MSRK) GeoAI framework combining a multiscale local singularity index, multitask Random Forest trend modelling, and residual kriging for precision-agriculture mapping of soil NPK, crop stress, and yield. 
|
|
STADOS: Spectral-Temporal Adaptive Dark Object Subtraction A five-stage radiometric harmonization framework that generalizes classical dark-object subtraction into a multi-object, reliability-weighted, iteratively self-consistent correction for multi-temporal Sentinel-2 NDVI monitoring, applied to corn parcels near Coimbra, Portugal. 
|
|
Zaria Crop ET and Irrigation DSS A modular Python pipeline with a Tkinter desktop GUI for reference/crop evapotranspiration, soil-water balance, irrigation scheduling, and full farm-report generation for five crops grown around Zaria, Nigeria. 
Crop growth-stage simulation with per-stage water accounting. |
|
GIS-Based Livestock Slurry Suitability A multi-criteria GIS overlay combined with regression kriging and deep-learning suitability surfaces for livestock slurry application siting, case study in Tudela, Navarre, Spain. 
|
|
Integrated Dairy Processing Decision-Support Model Compares several prediction models, then uses linear programming and repeated random-scenario simulation to recommend which dairy product to make each day under uncertain milk quality and market prices. 
|
Python · GIS · Remote Sensing · Deep Learning · Computer Vision ·
Geostatistics · Google Earth Engine · Sentinel-1/2 · Precision Agriculture ·
Bayesian Methods · Decision Support Systems
Full list of languages, libraries, techniques and tools across all repositories: Software_Stack_All_Repositories.pdf