Consulting & Research

  • Spatial Data Science
  • Applied Physics
  • Mathematical Modelling


  • 2023: Working with Geospatial Data in Python | datacamp
  • 2023: Exploratory Data Analysis in Python | datacamp
  • 2023: Writing Efficient Python Code | datacamp
  • 2022: Object Oriented Programming in Python | datacamp
  • 2021: Getting started with TensorFlow 2 | Imperial College (Coursera)
  • 2021: Hyperparameter Optimization for ML | Soledad Galli (Udemy)
  • 2021: Feature Engineering | Kaggle.com
  • 2020: Automating QGIS 3.xx with Python | Udemy
  • 2020: Intro to Deep Learning with PyTorch | Udacity
  • 2020: QGIS 3.0 for GIS Professionals | Udemy
  • 2020: SQL for Data Science | UC Davis (Coursera)
  • 2020: Geographic Information Systems (GIS) Specialization | UC Davis (Coursera)
  • 2019: The Complete Introduction to OpenFOAM | Udemy
  • 2019: Introduction to Engineering Simulations (ANSYS) | Cornell (edX)
  • 2019: CFD Python, Lorena A. Barba group | Jupyter notebooks available on GitHub.
  • 2018: Machine learning with Python | IBM (Coursera)
  • 2018: Python for Research | Harvard University (edX)
  • 2014: Electrical System Grounding & Electromagnetic Interference Analysis | SES & Technologies
  • 2013: Integrating renewable energy into the power grid | CPD international
  • 2013: Principles of Economics for Scientists | California Institute of Technology (Coursera)
  • 2012: Power Factory Fundamentals | DigSilent Pacific
  • 2009: Advanced statistics | MIC quality, www.micquality.com
  • 2006: Recent advances in Si solar cells | WCPEC, Hawaii.
  • 2002: 3rd Eurosensors school on fundamentals of sensor science and technology | Eurosensors XVI
  • 2001: Presentation skills, and intellectual property management | Two-day CRC for uTech event, Melbourne
  • 93-2022:  French, Spanish, Mandarin | Self taught & emmersion

I'm available for new projects 2024.

I've made a separate site, wekaResearch.com, for my consulting business. That's been going well, I'm flexible as to future employment arrangements.

I prefer on-site work over remote, and I'm keen to travel or relocate for this. I'm an AUS & NZ citizen, also eligible for the BC PNP Program (Ca).

Additional Training
  • Machine learning
  • Artificial neural networks
  • Python
  • SQL
  • Geospatial analysis
  • GIS, webGIS, GIS programming
  • Sensor science and technology
  • Power systems analysis
  • Renewable energy integration
  • Silicon solar cells
  • Statistical process control
  • Version control
  • OpenFOAM
  • Economics
  • Intellectual property management
  • Site safety
  • Job safety analysis

Data Sources:

The images were part of a dataset of 621 aerial boat and ship images downloaded from Kaggle

Object Detection Deep Learning Model:

The model was a fork of YOLOv5, by Ultralytics. Training was performed with an 80/20 training/validation split, default hyperparameters and took two hours on an Nvidia GTX1060 GPU.

Green boxes are the ground truth labels, red boxes are model predictions, with modeled probablility in white text


Spatial Data Science

Spatial data is my most recent career adventure. It's a great direction for me; combining interests in conservation and geography with strong math, sensor-science, and programming skills.

My spatial & remote sensing skills include:

  • Machine learning & computer vision
  • Spatial statistics
  • Optical physics and sensor science
  • Fundamentals of spatial data types & structures
  • Vector & raster analysis with a wide variety of processing tools
  • GIS programming, with Python, JavaScript & Spatial SQL
  • Competence with ESRI & QGIS eco-systems
  • Web GIS
Aotearoa - New Zealand

Data Sources:

Bathymetry data sourced from NIWA: NZ Bathymetry 250m Imagery/Raster layer, Niwa Open Data Licence

DEM sourced from LRIS 25m DEM 2010: North Island, South Island, Landcare Data Use License.


Bathymetry data split into two layers: Top layer for stretched color symbology, 50% transparent. Lower layer is a hillshade, light source 315 deg azimuth, 30 deg altitude, z-factor: 10

DEM split into two layers: Streched color symbology layer downsampled to 250m, 50% transparent. Hillshade layer downsampled to 1000m, light source 315 deg azimuth, 30 deg altitude, z-factor: 10

13 September 2020, using ArcMap 10.7.1 on a Dell E6430.

Geographic Coordinate System: GCS_WGS_1984
Datum: WGS84
Projection: Mercator


Energy Systems

Four years as a full time power systems engineer and six as a solar cell scientist. I've kept up skills in power systems analysis, recently performing dynamic modelling on solar & geothermal plants with my old team at Electronet.


Microfabrication, MEMS & Solar Cells

I spent ten years overseas doing research in Micro-Electro-Mechanical-Systems (MEMS), photonics & PV. I'd consider relocating again for the right project.


Technology Transfer
Process Integration
Machine Learning
Connection Studies
Earthing Systems
Model Validation Fieldwork

For my PhD I developed a process to make compact photonic integrated circuits with Silicon-On-Insulator (SOI) waveguides. By using anisotropic wet etching I made smooth & well aligned internally reflecting corner surfaces.

I also solved some intermediate problems that were interesting in their own right such as better control over the etch planes. I found it was possible to switch from 45 degree {110} planes to vertical {100} planes with a KOH + IPA solution, by selecting the right temperature and KOH concentration.

I lacked the equipment to make an LPCVD nitride, so I investigated alternative etch masks. I tried various metals and metal oxides and found that both titianium and TiO2 worked well for deep silicon etching. It was more difficult to pattern TiO2, but this could be performed prior to oxidation.

  • 2006 ‘The use of titanium and titanium dioxide as masks for deep silicon etching’, O. Powell, D. Sweatman, H.B. Harrison, Smart materials and structures, vol. S81-86.
  • 2004 ‘Fabrication of Micro-Mirrors in Silicon Optical Waveguides’ My PhD thesis.
  • 2002 ‘Single-mode condition of silicon rib waveguides’, O. Powell, IEEE J. Lightwave Technology, vol. 20, pp. 1851-1855. See also the erratum in vol. 21, p. 868.
  • 2001 ‘Anisotropic etching of {100} and {110} planes in (100) silicon’, O. Powell, H.B. Harrison, J. Micromechanics and Microengineering, vol. 11, pp. 217-220.
  • 2004 ‘Integrated waveguides for optical interconnects’, D. Sweatman, O. Powell, S. Francis, Electronics Packaging Technology Conference. Re-published in Circuit World, vol. 32, pp. 3-7 (2006).
  • 2003 ‘The use of titanium and titanium dioxide as masks for deep silicon etching’, O. Powell, H. B. Harrison, D. Sweatman, SPIE conference on Device and Process Technologies for MEMS, Microelectronics, and Photonics III. Proc. SPIE, vol. 5276.
  • 2002 ‘Crystal alignment method for small SOI wafer pieces’, O. Powell, H. B. Harrison, Eurosensors XVI, pp. 209-211, Prague.
  • 2000 ‘Techniques for micromachining multilayered structures in silicon’, O. Powell, D. Sweatman, H.B. Harrison, IEEE conference on optoelectronic and microelectronic materials and devices (COMMAD 2000), 407-420, Melbourne.

My work in PV was more like MEMS processing than traditional PV. We produced Si solar cells by etching extremely deep narrow trenches in 2mm thick (110) wafers. This threw up all manner or interesting challenges I helped to address.

  • Prevention of mechanical stiction
  • Good exchange of liquid and gas solutions
  • Controlability of plasma processes
  • Depletion of chemicals due to the high surface areas
  • Through wafer uniformity of deposition and etching.
  • Removal of sacrificial dielectrics
  • Reliability of pre-furnace cleaning processes
1 /8
Internally reflecting corner mirror for a SOI large cross section waveguide.
2 / 8
The same waveguide corner looking from the top.
3 / 8
Six identical mesas in silicon, with the same orientation but changing etch chemistry.
4 / 8
Demonstrating the change of behaviour with changing chemistry, on a single multi-level structure.
5 / 8
Titanium Oxide as an alkaline etch mask.
6 / 8
The world's smallest sail number?
7 / 8
A random heap of SLIVER cells. No the real process doesn't have a step like this!
8 / 8
An interesting image of the side-wall. The surface features are greatly exadurated to show etching effects.

From 2005 through to late 2009 I was working on SLIVER solar cells in Adelaide, South Australia. The original process having been developed at the Australia National University.

Our $25M purpose built plant in Adelaide was nominally for pilot production. In reality we had to refine the process substantially before it was manufacturable. We learned many expensive but useful lessons along the way.

Once ready for production we had quite a different set of gear, and had shifted from 6 inch FZ to 8 inch CZ. We formed a joint venture with Micron Technology, USA and modified a de-comissioned 8 inch memory fab in Boise, Idaho to ramp up the cell process. The module assembly process was to revive an old HP manufacturing facility, in nearby Nampa.

At this point my focus shifted from process development to technology transfer, mentoring and bringing the team in Idaho up to speed with the process, and PV fundamentals more generally.

In parallel to relocating the manufacturing process I was involved in performance optimisation, reliability testing and qualification of the modules.

We independently found the problem that later became known as potential induced degradation (PID) and developed strategies to minimise it.

Initially it wasn't clear if this issue was unique to our cells or something industry wide. SLIVER cells were using a surface passivation method with I developed involving an LPCVD nitride anti-reflection coating.

Process integration was a key part of my role at Origin Energy. Getting from incoming wafers to working solar cells took in the order of 100 steps (vs 25 for a N-IBC cell). This could potentially be justified as the area final area covered by the module was substantially larger than the area of the original wafer.

What made this all so creative and challenging was that we were making working solar cells on a completely three dimensional microstructures, with the active cell surfaces down extremely deep grooves.

The majority of the steps would have looked quite conventional in any MEMS or IC process, the key challenge was fitting them together in the simple way with high reliability and yield. Over the six years I developed several new working processes, each slightly better than the previous one.

As the fab I was using had been purpose-built for the original process from the ANU, many of the new ideas had to be performed externally. I collaborated with several external institutions. My highly fragile wafers travelled about the world, returning to South Australia for critical steps.

1 /6
A single SLIVER cell, flexible monocrystalline silicon, fully symmetrical and bifacial.
2 / 6
A 150mm wafer of unharvested cells, sitting on top of a partially completed module.
3 / 6
Conceptual cross section of a wafer.
4 / 6
Schematic diagram of the PV module layout.
5 / 6
An SEM image of the textured cell surface.
6 / 6
A big array of our modules sitting on top of a building at the Australian National University.
  • USPTO Patent Application 20120266949 for invention of a method to form one of the electrical contacts on SLIVER cells.
  • 2007 ‘Design of SLIVER cells to optimise performance and reliability’, M. Kerr, M. Stocks, J. Seymour, P. McCaffrey, C. Balla, O. Powell, P. Charles, D. Gordeev, N. Tothill, D. Gordeev, N. Tothill, P. Mackey. 22nd EU-PVSEC, Milan, Italy.
  • 2006 ‘Random isotropic texturing of SLIVER cells’, O. Powell, M. Stocks, D. Questiaux, M. Kerr, P. Verlinden. WCPEC, Hawaii
  • 2005 ‘High performance characteristics of SLIVER silicon solar cells’, M. Stocks, P. Verlinden, D.Gordeev, M. Stuckings, A. Lin, O. Powell, M. Kerr and P. Mackey. Invited paper, 20th Euro. PVSC, Barcelona.

Ninth Example

This is on the back of the last page, won't be seen

Eighth Example

If we hit that bullseye, the rest of the dominoes will fall like a house of cards. Checkmate.

Seventh Example

My instinct is to hide in this barrel, like the wily fish.

Sixth Example

I put it together mostly from memory, reliving some more of my long distance cycling adventures through Google Maps.

You can zoom and pan with the mouse.

Fifth Example

The lines show the routes different teams took for the 2007 "Beijing to Paris Carfree" expedition. I created this event with Ting Wú. 16 riders joined from Aotearoa-NZ, Taiwan, Australia, Malaysia & Germany. We were celebrating car-free mobility on the 100th anniversary of the original Peking-Paris Challenge.

Data Sources:
The route polylines I coded by hand.
Political borders are from ESRI.
Bathymetry & elevation from SRTM30_PLUS v8, eAtlas, various original sources.
Geographic Coordinate System: GCS_WGS_1984
Datum: WGS84
Projection: Azimuth Orthographic

Fourth Example

Raster analysis to answer the question of how to sequester the equaivalent of the billion trees program (assuming it was all Pinus Radiata) by planting Aotearoa-NZ natives on river margins.

I only counted river margins going through existing farmland. The answer I came up with was 26m. But there are a lot of caveats to this. Starting with a lack of reliable data for carbon sequestration by ANZ natives (any of them!).

I initially tried this using vector methods but the processing requirements (RAM in particular) using ArcMap were too much for my old computer.

I'll make this into a zoomable web-map some time

Third Example

This one was an air quality study looking at ozone measurements in California and comparing with elevation and demographic data by county.

Second Example

This is the same ast the last one, but in a 3D view using arc scene, laying Sentinel-2 imagery over a TIN

Table of contents

This is where the html text should go for right hand material

Test link here

First Example

On the right is a map I generated showing streamlines for the brook santuary water catchment, generated from 1m LiDAR, with the stream line segment width correlated to the upstream catchment area. It was based on an earlier class exercise. I've since done the same thing with QGIS, and automated the process in Python.

GIS Portfolio
Olly Powell
10 October 2020


My geospatial journey started with an awesome Coursera specialisation in GIS with the University of California. Since then I've taken more of a Python-first approach with QGIS and open source libraries. I've also been putting a lot of effort into deep learning based machine vision.

My optical physics and personal interests lean most naturally towards remote sensing and conservation GIS. However I'm also curious about applications in built infrastructure and clean energy resource modelling.

I'm still tweaking the format and coding for this booklet. I'm using a mix of JavaScript and CSS3 animations. You can navigate with arrow keys or the buttons.

1 /9

Olly's GIS Portfolio

(mobile version)

My geospatial journey started with an awesome Coursera specialisation in GIS with the University of California. Since then I've taken more of a Python-first approach with QGIS and open source libraries. I've also been putting a lot of effort into deep learning based machine vision.

My optical physics and personal interests lean most naturally towards remote sensing and conservation GIS. However I'm also curious about applications in built infrastructure and clean energy resource modelling.

I'm still working on this. You can navigate with the arrow buttons above.
2 / 9
Brook Sanctuary water catchment
3 / 9
Same thing, cool graphics from a hillshade and Sentinel-2 imagery laid over a TIN
4 / 9
An air quality study in California.
5 / 9
Raster analysis of the Billion Trees program planted on river margins.
6 / 9
Routes of the 2007 Beijing-Paris Carfree Expedition.
7 / 9
I am the man with no name. Zapp Brannigan, at your service.
8 / 9
My instinct is to hide in this barrel, like the wily fish.
9 / 9
If we hit that bullseye, the rest of the dominoes will fall like a house of cards. Checkmate.

I've been practicing applied machine learning since 2018, though on-line courses, Kaggle and my own or consulting projects.

I've constructed predictive models using both tabular data and unstructured data, including acoustic, time-series, and image data.

I can build neural networks from scratch or use pre-trained models with PyTorch or with TensorFlow & Keras.

I've been working on object detection and image classification problems for aquaculture, conservation and remote sensing applications. Clients so far have included SnapIT, The Department of Conservation (NZ), and GFA Group (DE).

This was from a recent Kaggle comp doing object detection and tracking of Crown-of-thorns starfish from video footage of the great barrier reef.

My biggest project as a power systems engineer was a connection study for the Ngatamariki Geothermal Power Station. I had no previous experience in power systems, so it was a nice challenge.

Ngatamariki is a 100MW binary cycle plant, developed by Ormat Energy. They are fairly new to New Zealand's relatively small and unusual power system. A detailed model for load control had never been developed for Transpower.

I made the model with Matlab-Simulink from first principles and schematics of the plant. Later we validated and tuned the model, and associated PID settings during the initial startup testing of the plant.

Some more routine parts of this project were to develop a model and recommend settings for the AVR. Also to model the many VSD's in the plant to ensure the plant would meet Transpower's standards for harmonic loads.

Much of the knowledge developed with Ngatamariki I subsequently applied to a smaller 30MW plant, now called TOPP1.

I did some early feasibility work for proposed small hydro schemes and wind farms. This was in the early 2010's, after the Key government had killed off conditions for new renewables development, so none of the projects got built at the time but they were still quite interesting.

The main tasks were doing load-flow studies with PowerFactory (PF) or ETAP. I was looking at issues like reactive power control, and conductor sizing. The proposed locations were generally quite remote, and proposed to be embedded in relatively "weak" distribution networks, typical in rural New Zealand.

More recently I upgraded the TOPP1 load-frequency control model in PF, and built a PF model for the Epuron Solar Farm near Alice Springs.

At Mitton Electronet I developed models of high voltage earthing systems using the specialist electromagnetic modelling software CDEGS.

This role encompassed a variety of electomagnetic modelling tasks, from design of new infrastructure through to periodic reveiws of existing plant.

Some more noteable projects included:

  • Design, modelling and testing of the earthing system for the Ngatamiriki geothermal power station
  • Review and testing of the Tiwai Point alminium smelter
  • Earthing review and testing of the Mahinerangi wind-farm
  • Earthing review of the Kizildere II Geothermal power station, Turkey
  • Soil testing and modelling for the proposed HVDC system from Adelaide to the York Peninsula
  • Earthing review, testing and modelling for the NZ HVDC pole-3 upgrade
  • An electromagnetic interference study in Cessnock, NSW
  • Transient modelling of gas piplines in various locations through the central North Island

For model validation purposes, or routine safety inspections, I injection tested the earthing systems of high voltage power infrastructure throughout New Zealand and Australia for various energy utilities.

This process usually involved injecting a small 58Hz sample current into a remote location and measuring the voltage rise on the earthing grid and associated infrastructure, with respect to a reference in a second remote location.

Where practical we used phone or out of service power circuits for voltage refrences and injection currents. The rest of the time we rolled out cables by hand.

Some of the locations I did injection testing or soil resistivity testing of substations and other HV infrastructure.


If you could please fill in the form below I'll get back to you as soon as I can.

Kia kaha

This is the Remarkables range, in Otago. The left part of the image derived from 1m LiDAR elevation data and hydrology vector layers, the right hand side is from 0.3m aerial imagery

Data Sources:

River centrelines, river polygons, lake polygons, LiDAR tiles, Aerial Imagery are all from from the LINZ Data Service, Creative Commons Attribution 4.0 International

Projection: NZGD2000 / New Zealand Transverse Mercator 2000
Datum: New Zealand Geodetic Datum 2000