Kenneth Mayfield
Project OwnerI design & build detailed immersive training environments + intelligent, conversational AI 3D avatars, & predictive machine learning models. PM, planner. https://www.linkedin.com/in/xyriskenn/
Galiano Island’s warm-summer Mediterranean climate & rich biodiversity are stressed, declining from climate change & past land use. We propose a predictive model to forecast how environmental changes affect flora & fauna, based on scientific research & data by Island resident Andrew Simon, CanGov climate models, map & weather data, & thousands of research-grade anonymized species observations. We'll also build an agentic simulation of flora/fauna archetypes to simulate animal & plant behavioural adaptations as services on the SingularityNET Marketplace. Our results will directly aid Galiano Island's deep ecological restoration, educate, & improve policy change across these Gulf Islands.
New AI service
Service will output predictive data of environmental changes with plant/animal clustering based on Mr. Simon's dataset CanGov climate models weather data map data & thousands of species observations. Our team's ML developer accomplished 95% accuracy for our Flood Prediction Model now on the SingularityNET Marketplace.
Timestamp and latitude/longitude decimal format or virtual world coordinates translated to geo-synced real-world locations of Galiano Island soil moisture measurement.
Area-specific predictions of species presence & abundance. Soil moisture data. Environmental data.
New AI service
This service intends to simulate interactions useful to observing simulated flora & fauna behaviour in changing environmental conditions on Galiano Island. We anticipate representing diurnal activities interspecies interactions (e.g. pollination hunting) population trends & behavioural adaptations.
Species family iDs (plant insect mammal bird) time of year time of day location soil moisture level temperature.
Periodic updates representing agentic animal behavioural modes travel agent id species name family start location end location.
Galiano Island is a jewel of biodiversity in the Gulf Islands region of British Columbia, Canada. This is traditional territory of the First Nations peoples, Penelakut. Hwlitsum, and Tsawwassen, the Hul'qumi'num-speaking people.
Majestic whales, cunning dolphins, Moonglow anemones. powerful eagles, silent owls, mercurial barn swallows, energetic minks, busy bumblebees and so many more forms of life are found here. Citizen and professional scientists have made millions of observations of more than ten thousand species in the Gulf Islands area.
On Galiano Island alone, there are more than 3,000 observed species.
Past agricultural & livestock activity drained wetlands & reduced native biodiversity. Community groups now work to restore historic wetlands, springs, streams and meadows.
Scientific Basis of Proposal Research
Please refer to my Milestone 3 report:
https://drive.google.com/drive/folders/1UAhSzwRK6MCTW6T1InhRWPbOLQVd0JvU?usp=sharing
Mr. Simon et al: “Phylogenetic restriction of plant invasion in drought-stressed environments: implications for insect-pollinated plant communities in water-limited ecosystems”
Full link: Research Gate
Dataset Diversitree R R code for ecological data analysis
Simon’s team put together a phylogeny of Galiano Island seed plants based on a strong source from GenBank sequence data (353,185 seed plants). 207 insect-pollinated plants (entomophilous) were sampled, with 173 species represented in Mr. Simon’s study area.
In Round 3 we cross-referenced Simon's study locations with iNaturalist.org community observations to build a "tree of life" framework for our project.
We propose these technical solutions:
Predictive model based on Mr. Simon's paper and dataset, map data, weather data, historical CanGov climate models & research-level observations recorded at iNaturalist.org
Agentic prototype simulation of generalized species family behaviour responding to model outputs
C# Unity scripts for the community to integrate new Marketplace services with real-time visualization.
We'd selected Mount Sutil for the dry (xeric) location and Rees' Field for the wetland (hydric) region. For a moderate (mesic) soil moisture gradient between these locations, I chose Morgan Road/Bluffs Park area.
Note about bats & data scarcity
Bats are an enigma on Galiano: populations & locations are not entirely understood. Two species observed: Mouse-eared bats (Genus Myotis) & Silver-haired bat (Lasionycteris noctivagans). Perhaps our agentic simulation will generate insights vital to the Island's ecology.
Our Round 3 Ideation filtered observations by area and species family.
For brevity I'll present all species together per each zone:
Xeric (dry) zone: Mt. Sutil (research grade observations only)
https://www.inaturalist.org/observations?nelat=48.872560728132356&nelng=-123.3747644281641&quality_grade=research&subview=map&swlat=48.86674563455329&swlng=-123.38214586737308&verifiable=any
Mesic (Moderate) zone: Morgan Road/Bluffs Park (research grade observations)
https://www.inaturalist.org/observations?nelat=48.881848702859195&nelng=-123.34260227790686&place_id=any&quality_grade=research&subview=map&swlat=48.87225239464197&swlng=-123.3614850293717&verifiable=any
Xeric (wetland) zone: Rees' Field (research grade observations)
https://www.inaturalist.org/observations?nelat=48.888214837038355&nelng=-123.3366471921101&quality_grade=research&subview=map&swlat=48.88279666135645&swlng=-123.34677521335034&verifiable=any
With these map=based species observations, recorded timing of life cycles, Mr. Simon's scientific records, and Canadian climate models, we'll develop an Island-wide predictive model and agentic simulation of species interactions using the model outputs.
Predictive Modeling: Develop a machine learning model that predicts the presence and abundance of plant and animal species based on environmental variables such as precipitation, soil moisture, and time.
Simulation API: Create an API that allows users to input different environmental parameters and receive predictions on how these changes will affect the biomes.
The success of the predictive model hinges on the quality and comprehensiveness of the data. The following data sources will be integrated:
Historical Climate Data: Precipitation, temperature, and soil moisture data from weather stations on Galiano Island and surrounding regions as well as Canadian government climate models.
Species Observation Data: Records of plant and animal species observed on the island, including seasonal abundance, distribution and life cycle from sources like iNaturalist.
Soil and Water Quality Data: Information on soil types, moisture levels, and historical changes in water tables across different regions of the island.
Land Use and Historical Maps: Maps showing historical land use changes, including areas drained for agriculture and grazing, to understand the long- term impacts on biodiversity.
Several machine learning models will be evaluated to determine the best approach for predicting species presence and abundance:
Random Forests: For capturing complex interactions between environmental variables and species data.
Gradient Boosting Machines (GBMs): For robust prediction with interpretability, allowing stakeholders to understand key drivers of change.
Neural Networks: For handling large datasets with non-linear relationships and for capturing seasonality effects over time.
Geospatial Models: Incorporating spatial data to model species distribution patterns across different biomes.
The selected model will be trained using cross-validation techniques to ensure generalization and prevent overfitting. Hyperparameter tuning will be performed to optimize model performance.
The model’s performance will be evaluated using metrics such as:
Accuracy: Correct prediction of species presence or absence.
F1-Score: Balance between precision and recall for species prediction.
Area Under the ROC Curve (AUC-ROC): For assessing the model's ability to discriminate between different species based on environmental conditions.
RMSE (Root Mean Squared Error): For evaluating the accuracy of abundance predictions.
An API will be developed to expose the model's predictions in a user-friendly manner. Users will be able to:
Input environmental parameters such as month, year, precipitation, and soil moisture.
Receive predictions on species presence and abundance across the island’s biomes.
The API will be designed to be scalable and integrate easily with the 3D simulation tool.
We propose a Python-based agentic simulation prototype to generate data for observing and analyzing flora and fauna behavior under changing environmental conditions on Galiano Island.
The service will quantify diurnal behaviors (sleep, forage, groom, sunning), inter-species interactions (pollination, feeding, hunting, hiding, reproduction), and population dynamics (spatial shifts, growth trends) across dry, moderate, & wetland environments. The initial phase focuses on generalized species groups, with later specialization.
Platform: Core implementation in Python using Mesa for rapid agent-based prototyping.
Processing: Ray and Python’s asyncio for distributed, asynchronous processing to scale to large agent numbers.
Communication: Paho-mqtt library for lightweight, real-time agent-to-agent and agent-to-environment messaging.
We will develop four agent families—plants, insects, birds, and small mammals—differentiated by diet (detritus, vegetarian, omnivore, carnivore) and, for insects, by locomotion (flying vs. crawling).
States: diurnal behaviors & needs (energy, hydration, safety, sleep, wake, reproduction).
Behaviors: Seeking (food, water), feeding, pollination, hiding, hunting, reproducing.
Perception: Environmental (light level, moisture, temperature) & other agent signals (movement, communication, gathering).
Reasoning: define a learning model to evaluate internal states against environmental conditions. Memory integration allows for adaptive goal prioritization, thus simulating “agentic” cognitive behavior.
Action: reasoning outcomes will drive agent actions. For example, a foraging bee may adjust its flight path based on time of day, temperature, prior experiences with food sources, and the presence of conspecifics.
Spatial Factors: local geospatial data (terrain, resource distribution) from our predictive model API will influence agentic reasoning & movement.
Dynamic Factors: Our predictive model for Galiano Island provides an API for user experimentation with environmental factors. We’ll evaluate developing a somewhat randomized autonomous change of API inputs for a self-running simulation mode.
Metrics & Logging: within scope, simulation events will be logged with Pandas and SQL.
Visualization: Initial visualizations will use Matplotlib/Seaborn, with more detailed 3D visualization supported by C# Unity scripts for community use.
Our project neither requires nor records any personally identifiable information.
IP addresses from identified adversarial nation states may be monitored solely to prevent misuse, bot attacks, and tampering.
Model API inputs are restricted to environmental parameters only. During training, we exclusively use species data from public domain sources or datasets provided with explicit permission; no personal or web-scraped data (e.g., author names) will be accessed.
Additionally, climate models from CanGov are public domain & anonymized.
Our objective is to equip community groups, scientists, and policymakers with tools to restore native ecosystems.
In Round 3 we researched in-depth the data, species observations, and mapping of Galiano Island then developed a machine model development strategy. For this milestone we will decide specific requirements for the following project components: Galiano Island Environmental predictive AI model: 1. Preparing Mr. Simon's dataset of soil conditions and effects on native & invasive plant flowering for machine model training. 2. Identify whether synthetic data is required 3. identify additional governmental plant databases useful to the model e.g. https://open.canada.ca/data/dataset/bb14c6bf-75f7-4ff2-b97e-689fa768905c 4. Examine the CanGov Global Climate model scenarios dataset for use in our project. https://climate-change.canada.ca/climate-data/#/cmip5-data 5. Identity weather data sources suitable to model training 5. Select iNaturalist.org API's observation data needed while avoiding use of any user-identifiable data e.g. observation author. 6. Estimate computational requirements / cloud computing costs for model training Agentic Species Simulation: 1. Define agentic species perception reasoning action and adaptability algorithms which will use agent-agent & environment-agent inputs 2. Define environmental simulation parameters (time of day soil moisture temperature time of year species clustering food and shelter) 3. Define cloud platform requirements & cost estimations Ethics and Privacy Terms and use data use & ethical use text will be decided.
PDF document containing research and parameter conclusions for project components’ development. Kanban board of anticipated tasks.
$3,000 USD
Ensure all topics are explored & organized well enough to begin work, with a positive peer review of our strategy.
1. Data collection: - Historical Climate Data: Precipitation temperature and soil moisture data from weather stations on Galiano Island and surrounding regions and CanGov climate models. - Species Observation Data: Records of plant and animal species observed on the island including seasonal abundance and distribution from sources like iNaturalist & CanGov. - Soil and Water Quality Data: Information on soil types moisture levels and historical changes in water tables across different regions of the island. - Land Use and Historical Maps: Maps showing historical land use changes including areas drained for agriculture and grazing to understand the long-term impacts on biodiversity from CanGov & Islands Trust sites as described in our Ideation document. 2. Feature Engineering Key features to be extracted and engineered from the data include: - Temporal Features: Month year seasonality patterns. - Environmental Variables: Precipitation levels temperature soil moisture content proximity to water bodies. - Species-Specific Features: Native vs. introduced species ecological roles (e.g. pollinators predators) habitat preferences. - Spatial Features: Geographic coordinates elevation distance from historical wetlands land cover types. Ethics and Privacy We will not be accessing any personal information nor web-scraping during data prep. Instead we will rely on vetted data sources like CanGov & Mr. Simon's dataset & arrange ethical API access with iNaturalist.org.
Record of collected and prepared data sources for evaluation (Github scripts documentation).
$8,000 USD
Readiness to begin model training.
1. Model Selection and Training Several machine learning models will be evaluated to determine the best approach for predicting species presence and abundance: - Random Forests: For capturing complex interactions between environmental variables and species data. - Gradient Boosting Machines (GBMs): For robust prediction with interpretability allowing stakeholders to understand key drivers of change. - Neural Networks: For handling large datasets with non-linear relationships and for capturing seasonality effects over time. - Geospatial Models: Incorporating spatial data to model species distribution patterns across different biomes. - The selected model will be trained using cross-validation techniques to ensure generalization and prevent overfitting. Hyperparameter tuning will be performed to optimize model performance. 2. Model Evaluation The model’s performance will be evaluated using metrics such as: - Accuracy: Correct prediction of species presence or absence. - F1-Score: Balance between precision and recall for species prediction. - Area Under the ROC Curve (AUC-ROC): For assessing the model's ability to discriminate between different species based on environmental conditions. - RMSE (Root Mean Squared Error): For evaluating the accuracy of abundance predictions. Ethics & Privacy In training the model we will compare results with historic climate data for accuracy. Terms of use will state this dataset is for educational purposes only.
Version 1 of our Galiano environmental prediction model available for evaluation VIA Github. Documentation of the predictive model training accuracy results.
$8,000 USD
75% or greater accuracy of model outputs predicting impact of environmental changes on Galiano Island's Biomes.
Develop a prototype simulation architecture in Python supporting later environmental complexity. 1. Environmental Specifications - Baseline water availability in three zones: dry moderate wetland - Temperature range - Time of day or night 1. Baseline Agents Generalized archetypes (with examples of actual species for clarity) shall include: - native plant species e.g. Tracheophyta (i.e. Red-Flowering Currant) - invasive plant species e.g. Tracheophyta (i.e. European Holly) - flying insect - pollinator e.g. Apidae (i.e. Bumblebee) - flying insect - predator e.g. Aeshnidae (i.e. Common Green Darner dragonfly) - crawling insect detritus eater e.g. Elateridae (i.e. Red-Bellied Click Beetle) - crawling insect predator e.g. Dytiscidae (i.e. Giant Green Water Beetle) - Bird seed eater e.g. Fringillidae (i.e. Purple Finch) - Bird insectivore e.g. Hirundinidae (i.e. Barn Swallow) - Raptor obligate carnivore e.g. Strigiformes (i.e. Barred Owl) - Small mammal vegetarian e.g. Sciuridae (i.e. Eastern Grey Squirrel) - Small mammal predator e.g. Mustelinae (i.e. Mink) 3. Prototype Simulation - Basic plant germination to flowering cycle - Pollinator foraging - Simple mutualistic interactions - Predation and avoidance for small predators 4. Data Collection and Visualization tools - Initial data logging framework tracking agent states and interactions - Basic visualization of ecosystem interactions Ethics & Privacy TOS will state sim is for educational purposes only.
Design documentation of agentic algorithms decision-making models and interaction rules. Basic functioning simulation prototype.
$8,000 USD
The prototype simulation is stable and reproducible on the development platform. Agent behaviours follow the rules for generalized species type.
We will integrate the Prediction Model outputs of milestones 2 & 3 into the agentic simulation to represent the changing environmental states that will influence agentic model behaviour such as time of day location soil moisture precipitation and temperature. Ethics & Privacy Terms of use will state this dataset is for educational purposes only and not meant to replace science-based observations nor determine policy changes.
Scripts and Github access demonstrating integrating AI model outputs integrated with the agentic simulation handler. Documentation of changes and video record of agentic behaviour changing with environmental inputs.
$8,000 USD
Observable changes in agentic behaviour as environmental inputs change.
Development of a small collection of utility scripts to integrate our prediction model services with a basic Unity template for community ecological real time visualization projects. - Data request - Data parsing - Conversion from Mercator projection latitude-longitude are extent to Unity world coordinates - Conversion from Unity world coordinates to Mercator projection latitude-longitude extents - User input VIA Unity UI of specific values to alter prediction model environmental inputs e.g. temperature time of year soil moisture level to aid experimentation Ethics & Privacy These scripts will only submit data inquiries to the API and will not include any personally identifiable information.
Collection of Unity-compatible C# utility scripts and an empty sample project which can demonstrate access to the model endpoints.
$2,000 USD
Functioning C# scripts and sample project which successfully connects to our proposed ML model API endpoint and returns data.
We will develop Unity C# scripts that synchronize the agentic simulation outputs with simple Unity objects. These objects will be 3D primitives representing the species families used in the agentic simulation. High fidelity of synchronization is not required between server and Unity scripts since a visualization is not fast-paced. We will focus on a limited frame-based server/client sync of object positions. Animation tags will define behavioural traits in data returned from the server-side API. For example: #rest #forage #run #search sleep etc. How these are implemented in a user's 3D simulation will be up to the user's choice. For example "forage" for a dragonfly is different from foraging behaviour of a bee. Ethics & Privacy These scripts will only submit data inquiries to the API and will not include any personally identifiable information. TOS will specify use is for educational use only. API interactions will be rate-limited for efficiency.
Collection of Unity C# scripts which connect to the agentic model simulation API and processes its data into motion commands and behavioural animation tags for Unity developers.
$3,500 USD
Demonstration of successful agentic simulation API usage in a basic Unity project.
Integration of our Galiano Island predictive model and agentic species simulation API endpoints with the SingularityNET marketplace.
Documentation of services integration with SingularityNET's marketplace.
$2,000 USD
Presence of this project's services on the SingularityNET marketplace.
The completed project will be presented online to the Galiano Island ecological restration groups. Ethics & Privacy The presentation record will focus on the presenter and content. TOS for educational purposes only will be announced VIA screen text.
Record of presentation focusing on the presentation and speaker.
$250 USD
Attendance of IMERSS stakeholders
This milestone will be used to compensate for cloud server for post-project delivery of the services and placement of any demonstration files on pixel-streaming services for a limited time.
Invoices of usage costs from cloud and pixel-streaming service providers.
$750 USD
Use of cloud and pixel streaming services indicating the project is live.
Reviews & Ratings
Please create account or login to write a review and rate.
Check back later by refreshing the page.
© 2024 Deep Funding
Join the Discussion (0)
Please create account or login to post comments.