Back to search

National Technical University of Athens (NTUA)

Big earth data and machine learning for sustainable and resilient agriculture

Abstract

dc:description

More than a tenth of the global population does not have access to sufficient quantities of affordable, nutritious food. At the same time, the food demand is expected to increase between 35-56\% over the period 2010-2050, which will require the intensification of agriculture. On the other hand, climate change calls for the employment of agricultural practices that will secure resilience and sustainability. There is, therefore, an urgent need for producing more, while changing our methods to account for future changes. In this regard, the large scale and frequent monitoring of agricultural land can provide significant insights for timely decision making based on evidence. Big streams of Earth images from satellites or other platforms (e.g., drones and mobile phones) are becoming increasingly available at low or no cost and with enhanced spatial and temporal resolution. This thesis recognizes the unprecedented opportunities offered by the high quality and open access Earth observation data of our times and introduces novel machine learning and big data methods to properly exploit them towards developing applications for sustainable and resilient agriculture. The thesis addresses three distinct thematic areas, i.e., the monitoring of the Common Agricultural Policy (CAP), the monitoring of food security and applications for smart and resilient agriculture. The methodological innovations of the developments related to the three thematic areas address the following issues: i) the processing of big Earth Observation (EO) data, ii) the scarcity of annotated data for machine learning model training and iii) the gap between machine learning outputs and actionable advice.The first contribution of this thesis is the Agriculture Monitoring Data Cube (ADC) that offers an automated, modular, end-to-end, cloud-based framework for handling big satellite data (Sentinel-1 and Sentinel-2) based on the Open Data Cube (ODC). ADC provides a set of powerful tools on top of the cube, including i) the generation of analysis-ready feature spaces of big satellite data to feed downstream machine learning tasks and ii) the support of Satellite Image Time-Series (SITS) analysis via services pertinent to the monitoring of the CAP, e.g., detecting trends and events, monitoring the growth status and more. The second contribution of this thesis is a scalable and transferable machine learning method for multi-crop classification. The method uses the farmers’ declarations, as part of their subsidy applications in the CAP, in order to train the crop classification model. The method is extended by applying semantic enrichment on the crop type maps, increasing the value of knowledge extracted towards making decisions in operational scenarios of the paying agencies of the CAP. Specifically, a smart sampling method was developed to select parcels of potential wrong declaration (i.e., the farmers do not cultivate what they declared). This method provides actionable advice to the inspectors of the CAP paying agencies, early in the year, according to their operating model requirements. Freely available satellite data, such as Sentinel-1 and Sentinel-2 data, used in the methods described above, cannot always provide confident crop type predictions in areas characterized by extended cloud coverage and/or small farm sizes. For this reason, ancillary sources of EO data are required. Towards this direction the space-to-ground paradigm is introduced, bringing together street-level and satellite images in an analysis-ready dataset, encouraging the community to experiment with fusion machine learning techniques for enhanced crop classification results. In this spirit, a preliminary late fusion approach was developed, where street-level image crop classification results were combined with the satellite based crop type maps. Another crop classification method is developed to classify paddy rice in South Korea. In this case, the focus is on weakly supervised learning as for South Korea there are not any available annotations. A pseudo-labeling approach is introduced using merely a confined number of labels that come from a small part of the country. Then the pseudo-labels are used to train a paddy rice classification model that generalizes to the entire country of South Korea. The nationwide classification required the processing of TBs of Sentinel images that would have been impossible in conventional machines. For this reason, a distributed implementation of the model was deployed using SPARK in a High Performance Data Analytics (HPDA) environment. Finally, this thesis introduces two methods for phenology estimation in cotton fields. This is important work towards timely farm interventions that will secure the quality and volume of the yield and even increase it. Phenology ground observations are scarce in time and space, for this reason both methods focus on utilizing only few labels. In the first method, a semi-supervised approach is developed that uses a handful of labels to generate thousands of pseudo-labels that in turn train multiple supervised crop phenology classification models. The second method takes it a step further and develops a fuzzy clustering approach that not only estimates phenology in an unsupervised way but additionally predicts the transitional states between phenological stages using the membership score of fuzzy c-means. The aforementioned contributions refer to detection methods, e.g., crop classification, phenology estimation. Although this work is important and addresses big data and machine learning issues associated with the large-scale and timely monitoring of agricultural land, there is still work to be done to reach actionable advice for the policy maker and the farmer. In this regard, causal and interpretable machine learning have been identified as key enablers to bridge this gap. Two methods have been developed to showcase preliminary results towards this direction. The first method uses interpretable machine learning to estimate the onset of pest harmfulness in cotton fields. The interpretability of the model allows for i) the rapid adoption of the application by the farmer and ii) the combination of the data-driven predictions with the empirical knowledge of the farmer, thus potentially increasing the value of the model outputs. The second method uses causal machine learning to assess agricultural land suitability for applying specific cultivation practices. In more detail, the heterogeneous impact of crop rotation and landscape crop diversity on Net Primary Productivity (NPP) was estimated, accounting for historical crop and environmental data. The results showed that the effect of crop rotation was insignificant, while landscape crop diversity had a small negative effect on NPP. Finally, considerable effect heterogeneity in space was observed for both practices. All in all, this thesis showed that big EO data are a powerful tool for the large-scale and timely monitoring of agricultural land towards food security and climate resilience. In this context, it was demonstrated how big data technologies such as data cubes, distributed learning, linked open data and semantic enrichment can be used to exploit the data deluge and extract knowledge to address real user needs. Furthermore, this thesis argues for the importance of semi-supervised and unsupervised machine learning models that circumvent the ever-present challenge of scarce annotations and thus allow for model generalization in space and time. Specifically, it is shown how merely few ground truth data are needed to generate high quality crop type maps and crop phenology estimations. Finally, this thesis argues there is considerable distance in value between model inferences and decision making in real-world scenarios and thereby showcases the power of causal and interpretable machine learning in bridging this gap.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Sitokonstantinou, Vasileios
  • Σιτοκωνσταντίνου, Βασίλειος

Subjects

dc:subject × 33

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/52604

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Sitokonstantinou, Vasileios; Σιτοκωνσταντίνου, Βασίλειος. Big earth data and machine learning for sustainable and resilient agriculture. National Technical University of Athens (NTUA), 2022. http://hdl.handle.net/10442/hedi/52604