Projects · SEFA

Project type

Remote Sensing Projects

Four panels compare image registration, with cyan and orange outlines becoming more closely aligned.

Software

Coregix

Creator and maintainer

Small positional offsets make it harder to compare images or combine data from different sensors. Coregix aligns a source raster with a reference raster using mutual-information optimization, with edge-based inputs for cross-sensor structural alignment. I created and maintain its Python API and command-line workflow, which write aligned GeoTIFFs while retaining geospatial metadata and multiple bands.

Code Website

NDVI time series comparing raw and harmonized Landsat 5, 7 and 8 observations.

Software

pixltsnorm

Creator and maintainer

Differences between sensors can interrupt otherwise continuous environmental records. pixltsnorm fits linear transformations from overlapping observations to bring numeric time series onto a common scale. It supports both pixel-level and pooled transformations, and can connect sensors through intermediate overlaps. I created and maintain the library following work on Landsat time-series normalization during my PhD.

Code Website

LiDAR points divided into vertical voxels beside equations for plant area density, plant area index and foliage height diversity.

Software

PyForestScan

Creator and maintainer

PyForestScan turns airborne lidar point clouds into measures of forest structure, including canopy height, plant area density, plant area index and foliage height diversity. It brings established calculations, large-dataset processing and GIS-ready outputs into Python, making it easier to build reproducible forest analyses. I developed the concept and initial software, with contributions from Benjamin Palsa Leamon, and maintain the library.

Code Website Software paper

Four panels show satellite and lidar inputs, the sam-stars workflow, classified tree crowns, and annotation efficiency compared with a Mask R-CNN baseline.

The sam-stars workflow connects multimodal inputs, individual crown delineation, and species classification, with an annotation-efficiency comparison against Mask R-CNN.

sam-stars graphical abstract (V2), supplied by Iosefa Percival. Image source · Used with permission

Software

samstars

Creator and maintainer

Separating individual tree crowns is difficult where neighboring canopies touch. samstars combines high-resolution imagery with lidar-derived canopy height and density layers to propose crowns and refine them into separate polygons. I created and maintain the Python library, including training and tiled segmentation workflows. The output supports subsequent crown-level analysis; users train a model bundle for their own data.

Code Documentation

Colored outlines distinguish classified tree crowns in RGB imagery and a matching canopy-height map.

Classified tree crowns over RGB imagery and a canopy-height model.

Classified tree-crown figure supplied by Iosefa Percival. Image source · Used with permission

Software

Object-Based Image Analysis in Python

Creator and maintainer

Object-based image analysis groups neighboring pixels into meaningful units before classifying them. I created and maintain obia to connect raster segmentation, feature extraction and classification in Python. It combines spectral and texture features with optional point-cloud measurements on the same objects, then exports segment polygons and attributes for GIS analysis.

Code Documentation

GIS Projects

A panoramic forest-trail photograph used as example input imagery in LandLensDB documentation.

Forest-trail example input photograph, Sefa (Joseph) Percival / LandLensDB. Image source · Used with permission

Software

LandLensDB

Creator and maintainer

Field and street-level photos are most useful when their locations and metadata stay connected to the images. LandLensDB brings these collections into a spatial workflow: extracting geotags, organizing image records, mapping them and storing them in PostgreSQL/PostGIS. I created the toolkit to make geotagged imagery easier to find, query and reuse alongside other geographic data.

Code Website

gwpcorMapper interface showing a Tokyo correlation map alongside a linked scatterplot.

gwpcorMapper contributors; basemap © CARTO and © OpenStreetMap contributors. Image source · GPL-3.0 · CARTO · OpenStreetMap

Software

gwpcorMapper

Creator and maintainer

A single correlation can hide very different relationships in different places. gwpcorMapper is an interactive R Shiny application that maps local correlation and partial correlation, which accounts for other measured variables. Linked maps and scatterplots let users inspect these patterns and explore analysis settings. I created and maintain the tool to make geographically weighted statistics easier to investigate visually.

Code Website Methods paper

A blue vehicle route crosses a street network between two marked locations in Hiroshima.

pgRouting workshop contributors; map data © OpenStreetMap contributors. Image source · CC-BY-SA-3.0 · OpenStreetMap

Software

vrpRouting

Co-creator and maintainer

Planning routes for a fleet means accounting for tasks, vehicle capacity and delivery schedules as well as distance. vrpRouting brings vehicle-routing algorithms into PostgreSQL, alongside PostGIS and pgRouting. Its VROOM integration supports jobs, pickups and deliveries, and time windows, so route planning can use data already held in a spatial database.

Developed with Vicky and the pgRouting team. I develop code and review pull requests.

Code Website

Six world maps compare geographically weighted correlation and partial correlation among NDVI, temperature, and precipitation.

Spatial relationships among NDVI, temperature, and precipitation in 2014, comparing geographically weighted correlation and partial correlation. Grey points indicate p-values above 0.05.

Figure 1 from Joseph Percival and Narumasa Tsutsumida, Geographically Weighted Partial Correlation for Spatial Analysis, GI_Forum (2017). Reproduced unchanged. Image source · CC BY-ND 4.0

Software

gwpcor

Creator and maintainer

GWpcor calculates geographically weighted correlation and partial correlation in R. It estimates relationships locally, with partial correlation accounting for the other variables included in the analysis. The package returns spatial results that can be mapped or examined further, helping analysts investigate associations that a single study-wide statistic can obscure. I created and maintain the package.

Code Website Methods paper

pgRouting workshop contributors; map data © OpenStreetMap contributors. Image source · CC-BY-SA-3.0 · OpenStreetMap

Software

pgRouting

Maintainer and Project Steering Committee member

pgRouting adds routing and network analysis to PostgreSQL/PostGIS. Analysts can use SQL to find shortest paths, calculate reachable areas and solve other graph problems using network data and costs stored in their database. Keeping the analysis alongside the data makes it easier to incorporate changing attributes and connect routing results to wider GIS workflows.

I review code and contribute to project governance.

Code Website

Ecology Projects

A hemispherical view looking up through a forest canopy, with patches of sky between leaves and branches.

Hemispherical canopy example, Iosefa Percival / skye. Image source · Used with permission

Software

skye

Creator and maintainer

Forest photographs offer a way to examine how much sky is visible through a canopy. skye is an experimental Python library that converts RGB 360-degree images into hemispherical views, detects sky regions and calculates sky view factor. I created and maintain it to explore photographic approaches to forest canopy analysis.

Code

Figure 2: a map of repeated fires in northwestern Madagascar, cumulative burned area over time, and fire severity by number of fires.

Percival et al. (2024), Fig. 2, PLOS ONE, CC BY 4.0. Landsat-8 imagery courtesy of the U.S. Geological Survey. Image source · CC BY 4.0

Applied Research

Fire and Forest Loss in North Western Madagascar

Doctoral research and ongoing collaboration

How does a dry forest recover after fire, and what happens when it burns again? In Ankarafantsika National Park, our study combined a Landsat fire history with GEDI lidar measurements of forest structure. We found limited early recovery and vulnerability to repeated burning in the studied forest, underscoring the importance of fire prevention. This work began during my PhD at Kyoto University and continues with research partners.

Read the study

Applied Research

Seascape Ecology in Mangroves

Master’s research

My master’s research at Kyoto University examined mangrove habitat structure, connectivity and fish communities in Samoa. It asked how the arrangement and connections among habitats relate to the fish communities they support, bringing a seascape perspective to coastal ecology.

3D Documentation Projects

Pōtini point-cloud survey showing the megalithic landform and surrounding trees.

Pōtini point-cloud survey, Living Monuments / Tiapapata Art Centre. Image source · Used with permission

Mapping Samoa’s Megalithic Structures

3D mapping and documentation with Tiapapata Art Centre, including the Tia at Pōtini in Saʻanapu, Samoa. Photogrammetric point-cloud surveys record the monument’s shape and its relationship to the surrounding landscape. The resulting 3D model provides another way to explore and document the site alongside the centre’s heritage work.

Explore Pōtini

Photograph of a hafted stone adze with a wooden handle and fibre lashings.

Hafted stone adze, Tiapapata Art Centre. Image source · Used with permission

3D Models of Stone Tools and Cultural Artefacts

3D modelling and documentation of stone tools and cultural artefacts with Tiapapata Art Centre. Photogrammetry uses overlapping photographs to record objects as three-dimensional models, making their form available for digital viewing and documentation. This work sits within the centre’s wider efforts to sustain and share Samoan cultural heritage.

Heritage project

Filter by tag