Joaquin Gajardo Castillo


Last Name

Gajardo Castillo

First Name

Joaquin

ORCID

0000-0001-9144-4384

Organisational unit

03894 - Walter, Achim / Walter, Achim

Filters

20254
Reset filters

Search Results

Publications1 - 4 of 4
  • Evaluating the role of training data origin for country-scale cropland mapping in data-scarce regions: A case study of Nigeria
    Item type: Journal Article
    Gajardo Castillo, Joaquin; Volpi, Michele; Onwude, Daniel; et al. (2025)
    ISPRS Open Journal of Photogrammetry and Remote Sensing
    Cropland maps are essential for remote sensing-based agricultural monitoring, providing timely insights about agricultural development without requiring extensive field surveys. While machine learning enables large-scale mapping, it relies on geo-referenced ground-truth data, which is time-consuming to collect, motivating efforts to integrate global datasets for mapping in data-scarce regions. A key challenge is understanding how the quantity, quality, and proximity of the training data to the target region influences model performance in regions with limited local ground truth. To address this, we evaluate the impact of combining global and local datasets for cropland mapping in Nigeria at 10 m resolution. We manually labelled 1,827 data points evenly distributed across Nigeria and leveraged the crowd-sourced Geowiki dataset, evaluating three subsets of it: Nigeria, Nigeria + neighbouring countries, and worldwide. Using Google Earth Engine (GEE), we extracted multi-source time series data from Sentinel-1, Sentinel-2, ERA5 climate, and a digital elevation model (DEM) and compared Random Forest (RF) classifiers with Long Short-Term Memory (LSTM) networks, including a lightweight multi-task learning variant (multi-headed LSTM), previously applied to cropland mapping in other regions. Our findings highlight the importance of local training data, which consistently improved performance, with accuracy gains up to 0.246 (RF) and 0.178 (LSTM). Models trained on Nigeria-only or regional datasets outperformed those trained on global data, except for the multi-headed LSTM, which uniquely benefited from global samples when local data was unavailable. A sensitivity analysis revealed that Sentinel-1, climate, and topographic data were particularly important, as their removal reduced accuracy by up to 0.154 and F1-score by 0.593. Handling class imbalance was also critical, with weighted loss functions improving accuracy by up to 0.071 for the single-headed LSTM. Our best-performing model, a single-headed LSTM trained on Nigeria-only data, achieved an F1-score of 0.814 and accuracy of 0.842, performing competitively with the best global land cover product and showing strong recall performance, a metric highly-relevant for food security applications. These results underscore the value of regionally focused training data, proper class imbalance handling, and multi-modal feature integration for improving cropland mapping in data-scarce regions. We release our data, source code, output maps, and an interactive GEE web application to facilitate further research.
  • The FIP 1.0 Data Set: Highly resolved annotated image time series of 4,000 wheat plots grown in 6 years
    Item type: Journal Article
    Roth, Lukas; Boss, Mike; Kirchgessner, Norbert; et al. (2025)
    GigaScience
    Background: Understanding genotype-environment interactions of plants is crucial for crop improvement, yet limited by the scarcity of quality phenotyping data. This Data Note presents the Field Phenotyping Platform 1.0 data set, a comprehensive resource for winter wheat research that combines imaging, trait, environmental, and genetic data. Findings: We provide time-series data for more than 4,000 wheat plots, including aligned high-resolution image sequences totaling more than 153,000 aligned images across 6 years. Measurement data for 8 key wheat traits are included-namely, canopy cover values, plant heights, wheat head counts, senescence ratings, heading date, final plant height, grain yield, and protein content. Genetic marker information and environmental data complement the time series. Data quality is demonstrated through heritability analyses and genomic prediction models, achieving accuracies aligned with previous research. Conclusions: This extensive data set offers opportunities for advancing crop modeling and phenotyping techniques, enabling researchers to develop novel approaches for understanding genotype-environment interactions, analyzing growth dynamics, and predicting crop performance. By making this resource publicly available, we aim to accelerate research in climate-adaptive agriculture and foster collaboration between plant science and machine learning communities.
  • Towards improving farmers livelihood in Nigeria using food price forecasting
    Item type: Journal Article
    Odion , Divinefavour; Gajardo Castillo, Joaquin; Defraeye , Thijs; et al. (2025)
    Journal of Agriculture and Food Research
    Nigeria's agricultural sector represents approximately 25 % of the country's overall GDP and is a major source of employment for its population. This sector is largely driven by smallholder farmers who grow fruits and vegetables on farms under 4 ha. Despite their significant contribution to food production in Nigeria, most smallholder farmers, approximately 70 %, live in poverty, earning less than $1.9 per day. One of the key factors contributing to this situation is a lack of access to market price information. Farmers currently rely only on historical prices observed in local markets to decide on when, what, where and the price to sell their produce. This can lead to suboptimal decisions, resulting in food loss and loss of potential income. To address this challenge, we developed a machine learning online pipeline. It utilizes a Random Forest model trained on historical monthly fresh produce prices and other macroeconomic factors like currency exchange rates for Nigeria, that are regularly scraped from the internet. We deployed our trained model through an open-source mobile application, Coldtivate. Our model accurately predicted market prices for crops such as tomatoes, onions, potatoes, and plantains in various Nigerian states. The prediction success rate of our model varied across the various states in Nigeria. It ranged from 1 % to 20 % in Mean Absolute Percentage Error (MAPE) for predictions up to 8 months ahead. When evaluated on a hold-out test set, it yielded an RMSE of ₦45.16. The average MAPE of our model, when considering state-time-commodity averages, is up to 5 % lower than other baseline models, including the benchmark rolling-average, CatBoost, XGBoost, and SARIMA. By detecting patterns and trends in food prices, farmers can use our tool to make more informed decisions about when and what to sell to optimize profit, thereby improving revenue. Furthermore, our model provides a foundation for future machine learning model development in food price forecasting in agrarian countries.
  • Wheat3DGS: In-Field 3D Reconstruction, Instance Segmentation and Phenotyping of Wheat Heads with Gaussian Splatting
    Item type: Conference Paper
    Zhang, Daiwei; Gajardo Castillo, Joaquin; Medic, Tomislav; et al. (2025)
    2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)
    Automated extraction of plant morphological traits is crucial for supporting crop breeding and agricultural management through high-throughput field phenotyping (HTFP). Solutions based on multi-view RGB images are attractive due to their scalability and affordability, enabling volumetric measurements that 2D approaches cannot directly capture. While advanced methods like Neural Radiance Fields (NeRFs) have shown promise, their application has been limited to counting or extracting traits from only a few plants or organs. Furthermore, accurately measuring complex structures like individual wheat heads-essential for studying crop yields-remains particularly challenging due to occlusions and the dense arrangement of crop canopies in field conditions. The recent development of 3D Gaussian Splatting (3DGS) offers a promising alternative for HTFP due to its high-quality reconstructions and explicit point-based representation. In this paper, we present Wheat3DGS, a novel approach that leverages 3DGS and the Segment Anything Model (SAM) for precise 3D instance segmentation and morphological measurement of hundreds of wheat heads automatically, representing the first application of 3DGS to HTFP. We validate the accuracy of wheat head extraction against high-resolution laser scan data, obtaining per-instance mean absolute percentage errors of 15.1%, 18.3%, and 40.2 % for length, width, and volume. We provide additional comparisons to NeRF-based approaches and traditional Muti-View Stereo (MVS), demonstrating superior results. Our approach enables rapid, non-destructive measurements of key yieldrelated traits at scale, with significant implications for accelerating crop breeding and improving our understanding of wheat development.
Publications1 - 4 of 4