Pierre-Philippe Mathieu’s research while affiliated with European Space Agency and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (40)


Designing (Not Only) Lunar Space Data Centers
  • Conference Paper

July 2024

·

10 Reads

·

·

·

[...]

·


ESA-ECMWF Report on recent progress and research directions in machine learning for Earth System observation and prediction
  • Article
  • Full-text available

June 2022

·

251 Reads

·

21 Citations

npj Climate and Atmospheric Science

This paper provides a short summary of the outcomes of the workshop on Machine Learning (ML) for Earth System Observation and Prediction (ESOP / ML4ESOP) organised by the European Space Agency (ESA) and the European Centre for Medium-Range Weather Forecasts (ECMWF) between 15 and 18 November 2021. The 4-days workshop had more than 30 speakers and 30 poster-presenters, attracting over 1100 registrations from 85 countries around the world. The workshop aimed to demonstrate where and how the fusion between traditional ESOP applications and ML methods has shown limitations, outstanding opportunities, and challenges based on the participant’s feedback. Future directions were also highlighted from all thematic areas that comprise the ML4ESOP domain.

Download

SELF-SUPERVISED LEARNING – A WAY TO MINIMIZE TIME AND EFFORT FOR PRECISION AGRICULTURE?

May 2022

·

135 Reads

·

11 Citations

The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences

Machine learning, satellites or local sensors are key factors for a sustainable and resource-saving optimisation of agriculture and proved its values for the management of agricultural land. Up to now, the main focus was on the enlargement of data which were evaluated by means of supervised learning methods. Nevertheless, the need for labels is also a limiting and time-consuming factor, while in contrast, ongoing technological development is already providing an ever-increasing amount of unlabeled data. Self-supervised learning (SSL) could overcome this limitation and incorporate existing unlabeled data. Therefore, a crop type data set was utilized to conduct experiments with SSL and compare it to supervised methods. A unique feature of our data set from 2016 to 2018 was a divergent climatological condition in 2018 that reduced yields and affected the spectral fingerprint of the plants. Our experiments focused on predicting 2018 using SLL without or a few labels to clarify whether new labels should be collected for an unknown year. Despite these challenging conditions, the results showed that SSL contributed to higher accuracies. We believe that the results will encourage further improvements in the field of precision farming, why the SSL framework and data will be published (Marszalek, 2021).


Figure 2. Data on average temperature for each year and an overview of monthly precipitation. Climatological data from March to July were used for the visualisation.
Figure 3. Examples of embeddings generated with principal component analysis (PCA). The six different colors represent the crop types. A) gives an overview of the raw data. B) shows the result after pre-training with SimSiam and Aug1. Here it is already possible to differ between clusters. C) shows the last step and the embeddings after fine-tuning with labels.
Self-supervised learning -- A way to minimize time and effort for precision agriculture?

May 2022

·

140 Reads

Machine learning, satellites or local sensors are key factors for a sustainable and resource-saving optimisation of agriculture and proved its values for the management of agricultural land. Up to now, the main focus was on the enlargement of data which were evaluated by means of supervised learning methods. Nevertheless, the need for labels is also a limiting and time-consuming factor, while in contrast, ongoing technological development is already providing an ever-increasing amount of unlabeled data. Self-supervised learning (SSL) could overcome this limitation and incorporate existing unlabeled data. Therefore, a crop type data set was utilized to conduct experiments with SSL and compare it to supervised methods. A unique feature of our data set from 2016 to 2018 was a divergent climatological condition in 2018 that reduced yields and affected the spectral fingerprint of the plants. Our experiments focused on predicting 2018 using SLL without or a few labels to clarify whether new labels should be collected for an unknown year. Despite these challenging conditions, the results showed that SSL contributed to higher accuracies. We believe that the results will encourage further improvements in the field of precision farming, why the SSL framework and data will be published (Marszalek, 2021).


ANALYSIS AND DETECTION OF WILDFIRES BY USING PRISMA HYPERSPECTRAL IMAGERY

June 2021

·

154 Reads

·

19 Citations

The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences

This paper deals with the analysis and detection of wildfires by using PRISMA imagery. Precursore IperSpettrale della Mis­sione Applicativa (Hyperspectral Precursor of the Application Mission, PRISMA) is a new hyperspectral mission by ASI (Agenzia Spaziale Italiana, Italian Space Agency) launched in 2019. This mission provides hyperspectral images with a spectral range of 0.4–2.5 µm and an average spectral resolution less than 10 nm. In this work, we used the PRISMA hypercube acquired during the Australian bushfires of December 2019 in New South Wales. The analysis of the image is presented considering the unique amount of information contained in the continuous spectral signature of the hypercube. The Carbon dioxide Continuum-Interpolated Band Ratio (CO2 CIBR), Hyperspectral Fire Detection Index (HFDI), and Normalized Burn Index (NBR) will be used to analyze the informative content of the image, along with the analysis of some specific visible, near-infrared and shortwave-infrared bands. A multiclass classification is presented by using a I-dimensional convolutional neural network (CNN), and the results will be com­pared with the ones given by a support vector machine classifier reported in literature. Finally, some preliminary results related to wildfire temperature estimation are presented.


Climate-based ensemble machine learning model to forecast dengue epidemics

January 2021

·

45 Reads

·

6 Citations

Dengue fever is one of the most common and rapidly spreading arboviral diseases in the world, with major public health and economic consequences in tropical and sub-tropical regions. Countries such as Peru, 17.143 cases of dengue were reported in 2019, where 81.4% of cases concentrated in five of the 25 departments. When predicting infectious disease outbreaks, it is crucial to model the long-term dependency in time series data. However, this is challenging when performed on a countrywide level since dengue incidence varies across administrative areas. Therefore, this study developed and applied a climate-based ensemble model using multiple machine learning (ML) approaches to forecast dengue incidence rate (DIR) by department. The ensemble combined the outputs from Long Short-Term Memory (LSTM) recurrent neural network and Categorical Boosting (CatBoost) methods to predict DIR one month ahead for each department in Peru. Monthly dengue cases stratified by Peruvian departments were analysed in conjunction with associated demographic, geographic, and satellite-based meteorological data for the period January 2010–December 2019. The results demonstrated that the ensemble model was able to forecast DIR in low-transmission departments, while the model was less able to detect sudden DIR peaks in some departments. Air temperature and wind components demonstrated to be the significant predictors for DIR predictions. This dengue forecast model is timely and can help local governments to implement effective control measures and mitigate the effects of the disease. This study advances the state-of-the-art of climate services for the public health sector, by informing what are the key climate factors responsible for triggering dengue transmission. Finally, this project summarises how important it is to perform collaborative work with complementary expertise from intergovernmental organizations and public health universities to advance knowledge and address societal challenges.



Two schematics that illustrate key motivations and guiding considerations for obs4MIPs. Left: depiction of the large and growing community of scientists undertaking the climate model analysis who are not necessarily experts in modeling or in the details of the observations. Right: depiction of the large number of quantities available from model output (e.g., CMIP) and obtained from satellite retrievals, highlighting that a much smaller subset fall in the intersection but are of greatest relevance to model evaluation.
(a) Key to interpretation of obs4MIPs dataset indicators and (b) an example of the search result display of the indicators and links to the “[Tech Note]” and “[Supplementary Data]” in the case of datasets that include those (e.g., TES ozone).
An illustration of a model–observation comparison using obs4MIPs datasets. This four panel figure shows December–January–February (DJF) climatological mean (1981–2005) results for an individual model (a), the CERES-4-0 EBAF dataset (b), a difference map of the two upper panels (c) and a difference between the CMIP5 multi-model-mean (MMM) and CERES observations (d). The averaging period of the CERES-4-0 DJF mean is 2005–2018 (units are Wm-2).
An illustration of a model–observation comparison using obs4MIPs datasets. Tropospheric ozone annual cycle calculated from CMIP5 rcp4.5 simulations and AURA-TES observations, averaged over the years 2006–2009, for the NH (a) and SH (b) mid-latitudes (35–60∘) at 250 hPa. The individual model simulations are represented by the different colored lines while AURA-TES is shown as the black line (with ±1σ shown in gray).
Observations for Model Intercomparison Project (Obs4MIPs): status for CMIP6

July 2020

·

233 Reads

·

31 Citations

The Observations for Model Intercomparison Project (Obs4MIPs) was initiated in 2010 to facilitate the use of observations in climate model evaluation and research, with a particular target being the Coupled Model Intercomparison Project (CMIP), a major initiative of the World Climate Research Programme (WCRP). To this end, Obs4MIPs (1) targets observed variables that can be compared to CMIP model variables; (2) utilizes dataset formatting specifications and metadata requirements closely aligned with CMIP model output; (3) provides brief technical documentation for each dataset, designed for nonexperts and tailored towards relevance for model evaluation, including information on uncertainty, dataset merits, and limitations; and (4) disseminates the data through the Earth System Grid Federation (ESGF) platforms, making the observations searchable and accessible via the same portals as the model output. Taken together, these characteristics of the organization and structure of obs4MIPs should entice a more diverse community of researchers to engage in the comparison of model output with observations and to contribute to a more comprehensive evaluation of the climate models. At present, the number of obs4MIPs datasets has grown to about 80; many are undergoing updates, with another 20 or so in preparation, and more than 100 are proposed and under consideration. A partial list of current global satellite-based datasets includes humidity and temperature profiles; a wide range of cloud and aerosol observations; ocean surface wind, temperature, height, and sea ice fraction; surface and top-of-atmosphere longwave and shortwave radiation; and ozone (O3), methane (CH4), and carbon dioxide (CO2) products. A partial list of proposed products expected to be useful in analyzing CMIP6 results includes the following: alternative products for the above quantities, additional products for ocean surface flux and chlorophyll products, a number of vegetation products (e.g., FAPAR, LAI, burned area fraction), ice sheet mass and height, carbon monoxide (CO), and nitrogen dioxide (NO2). While most existing obs4MIPs datasets consist of monthly-mean gridded data over the global domain, products with higher time resolution (e.g., daily) and/or regional products are now receiving more attention. Along with an increasing number of datasets, obs4MIPs has implemented a number of capability upgrades including (1) an updated obs4MIPs data specifications document that provides additional search facets and generally improves congruence with CMIP6 specifications for model datasets, (2) a set of six easily understood indicators that help guide users as to a dataset's maturity and suitability for application, and (3) an option to supply supplemental information about a dataset beyond what can be found in the standard metadata. With the maturation of the obs4MIPs framework, the dataset inclusion process, and the dataset formatting guidelines and resources, the scope of the observations being considered is expected to grow to include gridded in situ datasets as well as datasets with a regional focus, and the ultimate intent is to judiciously expand this scope to any observation dataset that has applicability for evaluation of the types of Earth system models used in CMIP.


Observations for Model Intercomparison Project (Obs4MIPs): Status for CMIP6

November 2019

·

152 Reads

·

2 Citations

Abstract. The Observations for Model Intercomparison Projects (Obs4MIPs) was initiated in 2010 to facilitate the use of observations in climate model evaluation and research, with a particular target being the Coupled Model Intercomparison Project (CMIP), a major initiative of the World Climate Research Programme (WCRP). To this end, Obs4MIPs: 1) targets observed variables that can be compared to CMIP model variables, 2) utilizes dataset formatting specifications and metadata requirements closely aligned with CMIP model output, 3) provides brief technical documentation for each dataset, designed for non-experts and tailored towards relevance for model evaluation, including information on uncertainty, dataset merits and limitations, and 4) disseminates the data through the Earth System Grid Federation (ESGF) platforms, making the observations searchable and accessible via the same portals as the model output. Taken together, these characteristics of the organization and structure of obs4MIPs should entice a more diverse community of researchers to engage in the comparison of model output with observations and to contribute to a more comprehensive evaluation of the climate models. At present, the number of obs4MIPs datasets has grown to about 80, many undergoing updates, with another 20 or so in preparation, and more than 100 proposed and under consideration. Current global satellite-based datasets include, but are not limited to, humidity and temperature profiles; a wide range of cloud and aerosol observations; ocean surface wind, temperature, height, and sea ice fraction; surface and top of atmosphere longwave and shortwave radiation; along with ozone (O3), methane (CH4) and carbon dioxide (CO2) products. Proposed products expected for inclusion for CMIP6 analysis include, but are not limited to, alternative products for the above quantities, along with additional products for ocean surface flux and chlorophyll products, a number of vegetation products (e.g. FAPAR, LAI, burnt area fraction), ice sheet mass and height, carbon monoxide (CO) and nitrogen dioxide (NO2). While most obs4MIPs datasets are delivered as monthly and global, greater emphasis is being places on products with higher time resolution (e.g. daily) and/or regional products. Along with an increasing number of datasets, obs4MIPs has implemented a number of capability upgrades including: 1) an updated obs4MIPs data specifications document that provides for additional search facets and generally improves congruence with CMIP6 specifications for model datasets, 2) a set of six easily understood indicators that help guide users as to a dataset’s maturity and suitability for application, and 3) an option to supply supplemental information about a dataset beyond what can be found in the standard metadata. With the maturation of the obs4MIPs framework, the dataset inclusion process, and the dataset formatting guidelines and resources, the scope of the observations being considered is expected to grow to include gridded in-situ datasets as well as datasets with a regional focus, and the ultimate intent is to judiciously expand this scope to any observation dataset that has applicability for evaluation of the types of Earth System models used in CMIP.


Fig. 1. (a) Initial, randomly initialized SOM. (b) The BMU and its neighbor are updated to become more similar to the presented training set.
Fig. 2. Burkina Faso dataset. (a) Sample Level-1α RGB product and (b) its 49-cluster version obtained through SOM clustering. Each cluster is associated to a color label.
Integration of SAR and GEOBIA for the analysis of time-series data

In this work, we present a new architecture for the analysis multitemporal SAR data combining classic synthetic aperture radar processing and geographical object-based image analysis. The architecture exploits the characteristics of the recently introduced RGB products of the Level-1α and Level-1β families, employing self-organizing map clustering and object-based image analysis aiming at the definition of opportune layers measuring scattering and geometric properties of candidate objects to classify. The obtained results have been compared with those given by literature and turned out to provide high degree of accuracy and negligible false alarms. The discussion is supported by an example concerning small reservoir mapping in semi-arid environment.


Citations (27)


... Consequently, hybrid models, amalgamating machine learning with physical modeling, are envisioned to hold greater promise for future development [4]. The research on physically driven machine learning is still in its nascent stages, and thus, the aforementioned reviews offer relatively limited insights into this aspect [5]. ...

Reference:

Artificial Intelligence as Key Enabler for Safeguarding the Marine Resources
ESA-ECMWF Report on recent progress and research directions in machine learning for Earth System observation and prediction

npj Climate and Atmospheric Science

... Their proposed contextself contrastive loss (CSCL), based on attention matrix computations, was especially effective at improving performance at object boundaries, which are often where semantic segmentation performs poorly. Marszalek et al. [327] relied on labeled data from past years to form pairs of samples from the same class. ...

SELF-SUPERVISED LEARNING – A WAY TO MINIMIZE TIME AND EFFORT FOR PRECISION AGRICULTURE?

The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences

... Advanced computational methods like machine learning (ML) and deep learning (DL) are well suited to multi-data analysis, particularly for datasets with challenging characteristics, such as multi-scalar variation. To date, most climate and health research utilising advanced computational methods focus on physical health outcomes like infectious disease (e.g., Boudreault et al., 2023;Schneider et al., 2021) rather than mental health. Of those studies that do consider mental health outcomes, most analyse global (e.g., Pizzulli et al., 2021) or single region (Fahim et al., 2022) aggregate data rather than multi-regional or small area data. ...

Climate-based ensemble machine learning model to forecast dengue epidemics
  • Citing Conference Paper
  • January 2021

... These labeled reference pixels constitute a sort of testing set to be compared with the labels assigned by the proposed methodology. As in [39,40], the described approach is common practice in the analysis of new PRISMA data. We provide confusion matrices and the following statistical metrics: ...

ANALYSIS AND DETECTION OF WILDFIRES BY USING PRISMA HYPERSPECTRAL IMAGERY

The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences

... Its introduction to the EO research community for information extraction from large satellite datasets is a relatively recent development. However, many of the applications utilising DL appear to revisit previous EO problems, albeit with the advantage of faster and more accurate solutions compared to traditional Machine Learning (ML) approaches [3,4]. ...

Machine Learning for Earth System Observation and Prediction
  • Citing Article
  • December 2020

Bulletin of the American Meteorological Society

... The amip simulations are forced with observations of sea surface temperatures, sea ice cover, and historical forcings (Eyring et al., 2016). GC3.1 is the configuration that underpinned the United Kingdom's contribution to CMIP6 Mulcahy et al., 2018;Walters et al., 2019). The most recent configuration (GC5.0) has not been documented yet but includes three changes affecting cloud that are particularly relevant to our analysis. ...

Observations for Model Intercomparison Project (Obs4MIPs): status for CMIP6

... Standardization of model output in a common format (Juckes et al., 2020) and publication of the CMIP model output on the Earth System Grid Federation (ESGF) facilitates multi-model evaluation and analysis (Balaji et al., 2018;Eyring et al., 2016a;Taylor et al., 2012). This effort is additionally supported by observations for the Model Intercomparison Project (obs4MIPs) which provides the community with access to CMIP-like datasets (in terms of variable definitions, temporal and spatial coordinates, time frequencies, and coverages) of satellite data (Ferraro et al., 2015;Teixeira et al., 2014;Waliser et al., 2019). The availability of observations and models in the same format strongly facilitates model evaluation and analysis. ...

Observations for Model Intercomparison Project (Obs4MIPs): Status for CMIP6

... Vehicle discussions did not adequately identify the prior approach of pixel-based motion estimates. Object segmentation and feature extraction has enhanced the ability to discriminate things in aerial images which has recently led to new developments [6][7]. Aerial views are needed for essential operations like disaster rescue operations and scope management in crops [8]. ...

Feature Extraction From Multitemporal SAR Images Using Selforganizing Map Clustering and Object-Based Image Analysis

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

... S EA surface temperature (SST) is a critical oceanic and climatic parameter [1]. It reflects the exchange of energy and material between the ocean and the atmosphere [2], [3]. Moreover, SST significantly influences climate change [4], [5], ocean circulation [6], [7], and ecosystems [8], [9]. ...

Review and assessment of latent and sensible heat flux accuracy over the global oceans
  • Citing Article
  • November 2017

Remote Sensing of Environment

... Indeed, efforts to address these limitations include merging rain gauge and satellite data, but gridded precipitation estimates remain inherently uncertain (Herold et al., 2016). Reanalysis may provide consistent climate estimates but are susceptible to observation uncertainties and model limitations (Parker, 2016;Buizza et al., 2018). MSWEP emerges as a particularly promising dataset as corroborated by recent studies (Beck et al., 2017(Beck et al., , 2019bXu et al., 2019;Turco et al., 2020). ...

Advancing Global & Regional Reanalyses

Bulletin of the American Meteorological Society