Data Provenance in Agriculture

Abstract

Soils are probably the most critical natural resource in Agriculture, and soils security represents a critical growing global issue. Soils experiments require vast amounts of high-quality data, are very hard to be reproduced, and there are few studies about data provenance of such tests. We present OpenSoils; it shares knowledge about data-centric soils experiments. OpenSoils is a provenance-oriented and lightweight e-infrastructure that collects, stores, describes, curates and, harmonizes various soil datasets.

Full text

Data Provenance in Agriculture
Sérgio Manuel Serra da Cruz
1(&)
, Marcos Bacis Ceddia
1
,
Renan Carvalho Tàvora Miranda
1
, Gabriel Rizzo
1
,
Filipe Klinger
1
, Renato Cerceau
1,2
, Ricardo Mesquita
4
,
Ricardo Cerceau
1
, Elton Carneiro Marinho
5
,
Eber Assis Schmitz
5
, Elaine Sigette
3
, and Pedro Vieira Cruz
1
1
Federal Rural University of Rio de Janeiro, Seropédica, RJ, Brazil
{serra,ceddia}@ufrrj.br
2
National Agency of Supplementary Health, Rio de Janeiro, RJ, Brazil
3
Federal Fluminense University, Volta Redonda, RJ, Brazil
4
SENAI-RJ, Rio de Janeiro, RJ, Brazil
5
Federal University of Rio de Janeiro, Cidade Universitária, RJ, Brazil
Abstract. Soils are probably the most critical natural resource in Agriculture,
and soils security represents a critical growing global issue. Soils experiments
require vast amounts of high-quality data, are very hard to be reproduced, and
there are few studies about data provenance of such tests. We present OpenSoils;
it shares knowledge about data-centric soils experiments. OpenSoils is a
provenance-oriented and lightweight e-infrastructure that collects, stores,
describes, curates and, harmonizes various soil datasets.
Keywords: Reproducibility Soil security Open data Data quality
Big data
1 Introduction
According to Food and Agriculture Organization (FAO)
1
, an agency of the United
Nations, the world’s population is expected to grow to about 9,6 billion by 2050. Thus,
there is widespread concern about the challenges to soil and food systems in meeting
the demand of populations for sufficient, affordable, and nutritious food. There are
similar concerns about meeting those challenges in ways that agriculture would benefit
hugely from common shared global agronomic data spaces.
The modern Agriculture is a data-centric interdisciplinary domain, with the inte-
gration of different subjects (from genomics to soil sciences), different scales (from
genes to geolocalisation) and, different markets (from local farmers to multinational
research teams). The ability to manage and explore these datasets is a crucial issue to
tackle the current sustainability challenges. A wide variety of datasets underpin
products and processes, which vary in size, complexity, structure, semantics, subject
matter and in how they are updated and used.
1
http://www.fao.org/about/what-we-do/en/.
©Springer Nature Switzerland AG 2018
K. Belhajjame et al. (Eds.): IPAW 2018, LNCS 11017, pp. 257–261, 2018.
https://doi.org/10.1007/978-3-319-98379-0_31

Soils are probably the most critical natural resource in Agriculture; they generate
environmental, health and socio-economic benefits that are vital to sustaining life on
Earth [1]. Soil experiments are indispensable sources of knowledge. Researchers
conduct several kinds of soils experiments which are characterized as long-term field
experiments (LTE) and short-term (in vitro and in silico) lab experiments (STE).
The LTE have been running for years in many parts of the world for the last 175-years-
old (e.g. Rothamsted) and need more time to execute the research procedures. On the
other hand, STE experiments can be performed in a few weeks or months and have the
potential to contribute to the improve LTE. Thus, it is essential to deliver to the
agronomic community a novel computing infrastructure that can share raw and curated
data and the provenance of STE and LTE and augment the reproducibility of soil
experiments. This paper presents a multi-layer e-infrastructure which bring innovations
to Soils Science using FAIR principles (Findable, Accessible, Interoperable, and
Reusable) [2], W3C PROV-DM
2
, open data and semantic web standards.
2 Experiments in Soils Science
Soil Science represents the area that studies the soil (and its properties) as a natural
resource, including soil formation, composition, classification, mapping, management
and use [1,3], these properties could be about physical, chemical, biological, and
fertility. Soils experiments are costly because the soils are incredibly diverse, and it is
necessary to treat them in a specific manner [3]. Any recommendation fits specific soil
and weather conditions. Besides, the soil properties have high spatial and time vari-
ability. Finally, changes in soil properties can often be proved and quantified only after
decades.
The LTE is essential in monitoring and understanding the changes in soil physics or
fertility occurring because of long-term agrotechnical operations. Their scientific and
practical value is immeasurable and keeps improving over the years. The information
about the soils use cannot be replaced by any other means [3]. Additionally, the STE
produced much of the data that built the sciences of soil physics, chemistry, and
biology [1,3]. STE often explore soil processes subject to change over decades, topics
such as aggregation, weathering, microbial activity, and soil fertility itself.
Although STE enriches soil models, most tend to be reductionist, isolating individual
components, and do not study the whole soil, with its high-order interactions that
become apparent only with time.
3 Open Soils
Data and provenance are the primary and permanent assets in OpenSoils (www.
opensoils.org). The architecture is an open, provenance-oriented, and lightweight
computational e-infrastructure which rely on layers to store, compute and share curated
2
https://www.w3.org/TR/prov-dm/.
258 S. M. S. da Cruz et al.

data of (STE and LTE) soils experiments [5]. Figure 1illustrates a conceptual view and
the flow of information in the architecture.
Layer 1 (End-users layer) - hosts on the OpenSoils Web portal; it collects soil data
directly from the LTE into OpenSoils database. The specialists can use mobile and web
applications (e.g., OpenSoils App, API and Wet Lab tools) to collect the data directly
in the fields (LTE experiments) and trace the route of each soil sample sent to chemistry
and physics laboratories to be analyzed. Usually, the morphological properties of the
soil are analyzed in situ by the specialists. OpenSoils app sends raw data to the cloud-
based database thought the API. After that, each soil sample is tagged and sent to
laboratories where the scientist does wet experiments and execute STE which evaluate
specific physic-chemical properties of each soil horizon and selected soil samples are
shipped to the UFRRJ’s soils museum.
Layer 2 (Services layer) - hosts soil models and data-centric scientific workflows
which ingest large amounts of legacy data and analyses the consistency of the incoming
data [3].
Layer 3 (Data layer) - stores and describes various soils datasets with metadata.
The internal structure supports a diversified degree of data granularity and uses a
database named OpenSoilsDB [5,6] which can store new curated soils data annotated
with provenance metadata. Much of the information needed to assure the data quality
and to allow researchers to reproduce STE experiments can be obtained by system-
atically capturing data provenance [4]. OpenSoilsDB can store provenance from ETL
workflows and scripts. ETL Workflow provenance consists of the record of the
derivation of a result (e.g., a soil experiment, an image, a map) by a computational
process represented as scientific workflows. Script provenance is obtained by running
the source code of scripts (e.g. R, Pyhton). OpenSoilsDB used W3C PROV-DM
recommendation to store provenance and was designed to support the FAIR principles
for scientific data management and data stewardship [2]. The principles ensure trans-
parency, reproducibility, and reusability of the experiments, facilitating data sharing
more systematically.
The database also supports the ingestion of legacy soils data imported through ETL
workflows. The layer can store scientific and governance data. Besides, to support open
data, we can use general-purpose data repositories (e.g., CKAN, Dataverse, DSpace,
Dryad, DataHub).
A specific thesaurus is used to add semantics and annotate soils data, allowing us to
link it as RDF triples in WikiData. The thesaurus used in the e-infrastructure is
Agrovoc [7], which is a SKOS-XL (Simple Knowledge Organization System eXten-
sion for Labels) concept scheme published as LOD (Linked Open Data). It covers
several areas of interest of the FAO including food, agriculture and, environment. This
thesaurus is used by researchers, librarians, and information managers for indexing,
retrieving, and organizing data in agricultural information systems.
Data management is not a target in itself, but a key conduit leading to knowledge
discovery and innovation in soil sciences. OpenSoilsDB database stores scientific and
governance data. The scientific data aims to serve high quality-assessed, georeferenced
soils profiles database to the Brazilian and international communities upon their
standardization and harmonization. Each soil profile description recorded in the data-
base has more than 43 entities, and 250 attributes to stores the soil properties and soil
Data Provenance in Agriculture 259

experiments (mineralogical, morphological, chemical, physical, and environmental
data). Furthermore, the database support data versioning and provenance; stores geo-
referenced soil data (text and images) about physic-chemical analytical data from each
horizon and soil samples analyzed in wet laboratories.
Data governance is an essential block in the knowledge base of information pro-
fessionals involved in supporting data-intensive research. Its adoption is advantageous
because it is a service based on standardized, repeatable processes, designed to enable
the data discovery and the transparency of data-related transformation processes.
Layer 4 (Governance layer) - hosts data licenses, re-use rights, analytical tools,
visualization and map generation services that can be connected to other software (e.g.,
ArcGIS, R or Jupyter) to generate analytical reports, prediction and raster maps.
Although received little attention in soils research communities, this layer is founda-
tional for soils security. The prime function of the layer is to improve and maintain the
citations and quality of the soils dataset; thus, to be successful at governance, quality
must be continuously measured, and the results continuously retrieved by the data and
services layers.
4 Concluding Remarks
Maintaining healthy soils is a key to modern agriculture. However, there is still much
computational work needed to be developed in soil sciences and more in-depth studies
to understand the role of data provenance in Agriculture. We introduced OpenSoils; it
is an e-infrastructure which share knowledge about STE and LTE in soils security using
FAIR, PROV, and semantic web approaches. The infrastructure is being developed and
aims to enhance reproducibility of experiments and deliver high-quality datasets,
knowledge and maps based on curated data.
Acknowledgments. This work was supported in part by the Brazilian agencies FNDE/MEC/
SESU, PIBIC/CNPq, Petrobras and CYTED networks BigDSSAgro and SmartLogistcs@IB.
Fig. 1. Overview of the conceptual data-flow in OpenSoils.
260 S. M. S. da Cruz et al.

References
1. Koch, A., et al.: Soil security: solving the global soil crisis. Glob. Policy 4(4), 434–441 (2013)
2. Wilkinson, M.D., et al.: The FAIR guiding principles for scientific data management and
stewardship. Sci. Data 3, 160018 (2016)
3. Körschens, M.: The importance of long-term field experiments for soil science and
environmental research –a review. Plant Soil Environ. 52,1–8 (2006)
4. Cruz, S.M.S., do Nascimento, J.A.P.: SisGExp: rethinking long-tail agronomic experiments.
In: Mattoso, M., Glavic, B. (eds.) IPAW 2016. LNCS, vol. 9672, pp. 214–217. Springer,
Cham (2016). https://doi.org/10.1007/978-3-319-40593-3_24
5. Cruz, S.M.S., et al.: Towards an e-infrastructure for open science in soils security. In: XII
Proceedings on Brazilian E-Science Workshop (BRESCI), pp. 59–66. SBC, Natal-RN (2018)
6. Rizzo, G.S.C., Ceddia, M.B., Cruz, S.M.S.: Banco de Dados Pedológico: Primeiros Estudos.
In: 5th Proceedings on Reunião Anual de Iniciação Científica (RAIC), pp. 1–2. UFRRJ,
Seropédica (2017). (in Portuguese)
7. Caracciolo, C., et al.: The AGROVOC linked dataset. Seman. Web 4(3), 341–348 (2013)
Data Provenance in Agriculture 261