Geoinformatics and Earth Observation Laboratory

Department of Geography ◆ Institute for CyberScience

Pennsylvania State University


The projects performed in the laboratory can be divided into three general categories:Environmental Hazards, Energy and CyberInfrastructure.

Environmental Hazards

The primary sponsor for these projects is the Office of Naval Research. Most of the project focus on the fusion of massive amounts of data from heterogeneous sources, including but not limited to satellite and aerial remote sensing data, atmospheric numerical models, ground sensors, crowd sourced measurements and social media.

Fusion of remote sensing, numerical models and social media data during emergencies

Primary Investigator: Elena Sava (PhD). External Collaborator: Alfred Kalyanapu, Tennessee Tech University.

The goal of this project is to generate real time maps by combining satellite, aerial and ground measurements with data harvested from social media to show the spatio-temporal characteristics of an emergency situation. The primary scenario investigated is in the context of flood extents. During a flood, remote sensing data provide an high resolution view of the event, however they are limited to snapshots in time due to atmospheric opacity and sensor revisit limitations. Numerical models can provide a continuous simulation, but they might be over- or under-predicting the flood extent. Using social media it is possible to "fill the gaps" in the remote sensing and numerical model data, and continuously monitor the spatio-temporal evolution of the flood.

Assessment of crowd sourced radiation measurement

Primary Investigator: Carolynne Hultquist (PhD). External Collaborator: Mark Coletti, Oak Ridge National Laboratory.

During the 2011 Fukushima crisis, accidental and controlled radioactive releases occurred over a period of three weeks. Government measurements are available that map the ground contamination due to these releases, primarily occurring in the area northwest of the Fukushima-Daiichi nuclear power plant. The Safecast project was created in the immediate aftermath, and thousands of portable radiation measurements devices were distributed to the general population. Since then, many more devices have been deployed globally (two of these Geiger counters are at our laboratory), which generated over 80 million geolocated crowdsourced radiation measurements. The project goal is the assessment of the quality of the Safecast data when compared to government measurements and to determine if distributed crowd sourced measurements provide reliable data during emergencies. Current work includes the collection of Safecast data when the sensor is attached to an unmanned aerial vehicle (UAV) to determine if it is possible to quickly gather reliable radiological data over large geographical areas.

Machine learning pixel classication of aerial imagery for flood detection

Primary Investigator: Elena Sava (PhD).

Thousands of aerial imagery are collected during emergencies using UAVs and airplanes. These images are usually acquired using regular RGB cameras, and lack IR channel. Detecting water from RGB image is a non-trivial problem. The goal of this project is to automatically classify thousands of images to identify the areas that contain water. These images can be fused with satellite and numerical model data to generate flood extent maps.

Using Google TensorFlow for feature extraction in aerial imagery

Primary Investigator: Liping Yang (PD). External Collaborator: Davide Del Vento, NCAR.

This project consists in performing massively parallel (e.g. 150,000 cores) runs of TensorFlow for feature extractions in aerial imagery. The goal is to automatically identify relevant targets (e.g. buildings) in thousand of geolocated images acquired using UAVs or airplanes in order to determine which might contain data that can be analyzed for damage. For example, a typical scenario is the collection of 100,000 images after a flood, and the need to identify how many of these images contain roads. Once all the images containing roads are identified, they are further processed to determine if they contain flooded areas. These images are then fused with other data (e.g. satellite or model based) to create flood extent maps.

Forecasting extreme heat hazards

Primary Investigator: Martina Calovi (PhD). External Collaborator: Olga Wilhemi, NCAR.

This project consists in generating heat index forecasts at a very high spatial resolution in urban areas. An algorithm was developed for the downscaling of a deterministic heat index forecast from a single mesoscale atmospheric model using past forecasts and a set of past measurements from crowd sourced meteorological stations. The heat index is then input in a hazard model that takes into account different characteristics of the built environment (e.g. building type, shading, access to cooling stations) to generate hazard maps.

Atmospheric source detection of unknown contaminants

Primary Investigator: Guido Cervone. External Collaborator: Sue Ellen Haupt NCAR.

This project consists in the identification of the characteristics of an unknown source of a potentially toxic gas detected from a limited number of ground sensors using atmospheric transport and dispersion models. An algorithm was developed that uses machine learning to iteratively perform numerical transport and dispersion simulations from tentative sources. The error between the simulated and observed concentration values is computed and it is used to rank the solutions. The process repeats until a solution is found with an error below a determined threshold.

Source Apportionment of CH4 in the Marcellus Shale Gas Area

Primary Investigator: Yanni Cao (MS - Graduated). External Collaborator: Ken Davis, PSU and Alan Taylor, PSU.

The project consists in identifying the release rate for different man-made and natural sources of methane distributed over Pennsylvania and New York using atmospheric models and genetic algorithms. An optimization process is run to minimize the error between observed and simulated concentrations, and to iteratively find different release rate that minimize the error.


The primary sponsors for these projects are National Science Foundation, US Army, and the National Center for Atmospheric Research (NCAR). Several of the projects are related to the use of the Analog Ensemble (AnEn) technique, which is being jointly developed at Penn State and at NCAR.

2D Analog Ensemble for short term atmospheric probabilistic forecasting

Primary Investigator: Laura Clemente (PhD). External Collaborator: Luca Delle Monache, Stefano Alessandrini, NCAR.

The goal of this project is to extend the current one dimensional AnEn algorithm to the spatial domain. By taking into account the spatial relationship between points, it is possible to better grasp the spatio-temporal relationship between physical variables. The goal is to generate probabilistic forecasts of atmospheric parameters that are consistent in space and time, and can be used for a variety of problems.

Analog Ensemble for long term forecasting of photovoltaic energy

Primary Investigator: Weiming Hu (PhD). External Collaborator: Michael Mann, PSU; Shanteanu Jha, Rutgers University.

The goal of this project is to use the AnEn algorithm to perform a dynamic downscaling of climate model runs to generate probabilistic forecasts of likely future conditions over the continental US at very high resolution. These probabilistic forecasts are then used as input to a photovoltaic array model that computes how much energy can be produced under different atmospheric conditions. The main goal is to determine which locations in the USA are most suitable for distributed solar power generation, taking into account different future climate regimens.

Analog Ensemble as a proxy measurement of predictability for the economic assessment of wind power generation

Primary Investigator: Mehdi Sharar (MS - Graduated). External Collaborator: Luca Delle Monache, NCAR.

The project consists in generating a measure of uncertainty for the predictability of wind speed at 80 meters at different locations in the continental US. This measure is used as input for an economic model that assess the feasibility of using wind farms to generate electricity. The AnEn algorithm is run for thousands of locations over the continental US, and its skill score is used as a probabilistic measure of uncertainty in predictability. Therefore, where the skill score is highest, it is easier to predict wind speed at 80 meters, and consequently the forecast of power produced is more reliable.

Integration of renewable energy sources in a microgrid

Primary Investigator: Gabriella Ferruzzi (PD). External Collaborator: Luca Delle Monache, NCAR.

This project consists in determining the optimal bidding strategy in a day-ahead market for a microgrid where a fraction of the power production comes from renewable sources (e.g. a photovoltaic array). It takes into account the uncertainty associated with the generation of electricity using renewable sources, the price of electricity at different times, and the risk of under- or over-producing electricity.

CyberInfrastructure Projects

These projects are aimed at building the CyberInfrastructure required by several projects.

Parallelization of TensorFlow using Apache Spark

Primary Investigator: Liping Yang (PD). External Collaborator: Davide Del Vento, NCAR.

The goal of this project is to run TensorFlow in a massively parallel environment using Apache Spark. The goal is to be able to run hundreds of thousands of concurrent TensorFlow applications on NCAR Yellowstone and Cayenne supercomputers.

Antenna to receive satellite data

Primary Investigator: Guido Cervone.

The goal of this project is to install a satellite receiving antenna on the roof of the Walker building to receive NOAA Polar satellite data in real time. The phase I (feasibility study) was already performed, and a temporary installation of the antenna was already performed by undergraduate students.

Sensor installation on UAVs

Primary Investigator: Guido Cervone.

The goal of this project is to create a common bus to install multiple sensors on commercial grade UAVs. We have purchased two UAVs and have modified them to carry a variety of sensors.