The team of more than 30 Omdena AI engineers collaborated with Norwegian cleantech startup Glint Solar to use AI to augment their solar site assessment tool for floating solar panels.
The project goal was to apply remote sensing techniques on satellite imagery to infer the depth of inland water bodies. This information can be added to GlintSolar´s solar site assessment tool to identify suitable sites and accelerate the green energy revolution.
As Global Warming continues to rise, there needs to be a way to tackle this problem. One such method is to install solar panels and harness the energy from the sun.
But due to the low availability of land to install solar panels, an increasing number of installers use inland water bodies and install floating panels utilizing the water surfaces. Using water bodies is especially attractive in places where the availability of land is low. Interesting locations could be drinking water reservoirs, water cleaning facilities, and hydropower reservoirs.
Source: Glint Solar
Bathymetry (depth map) and vertical water level variation over time are essential to evaluate when choosing the best locations for solar installations, as these have a significant impact on the number of panels that can be installed on a given area, as well as the overall cost Bathymetry, can be derived using multispectral satellite images, but all commonly used techniques require calibration data.
Furthermore, today’s techniques are susceptible to noise from varying bottom conditions and particles in the water (such as silt and algae). In inland water bodies, calibration data is seldom available, and the water often has a high degree of particles.
The project outcomes
The three main objectives of the project were as following:
Identifying the preprocessing steps to denoise the data for better model performance,
Building AI models to infer the depth of inland water bodies, and
Integrating the suitable model for the GlintSolar solar site assessment tool
The datasets collected in this challenge include Water Bodies Dataset (Sample_1), Bathybase Dataset, and Global Reservoir Dataset. Additionally, several other available datasets were also identified for GlintSolar to consider further.
Three preprocessing steps were made to denoise the satellite data: a general pipeline that preprocesses raster data from the Bathybase dataset to modeling ready format, a Cloud cover removal process using Sentinel 2 Level 2 images, and Algal Blooms Detection process using MODIS data. While these processes showed excellent results in clearing the noises in the satellite image data, the limitation is that the Algal Bloom Detection process was not integrated into the pipeline due to different image sources; however, a proof of concept was done for further development.
The models and deployment phase
Several models were tested, and the best-performing model was identified.
The deployment of the work put the code into modularized python scripts for production purposes, included all the required dependencies, and stored all the files on a DAGsHub repository. This challenge successfully identified parts of the preprocessing steps to denoise the satellite data and identified a well-performing model to predict the depth of water bodies from satellite images. The current modeling process is based on one lake/water body but can be developed in the future to accept multiple waterbody data for modeling. The result of this challenge provided a preliminary preprocessing and modeling pipeline as a minimal viable product that will be further developed and scaled up for integration into GlintSolar’s solar site assessment tool.
In partnership with Norwegian company Think Outside the team developed a solution to forecast the water availability coming from rivers and snowmelt into reservoir lakes as well as electricity prices to optimize the renewable energy production in Scandinavia.
Think Outside is a Norwegian company currently focused on providing clients with “constant access to accurate and reliable snow data … to make projections you can be confident in”.
They use radar systems to image snowbanks so that they can provide their clients within the hydropower energy industry with data about the density of snowpacks. The hydropower companies can use this data to make better predictions about the future volume of water that will be flowing into reservoirs due to snow melting and from this make more accurate predictions about the amount of energy they will be able to generate from this water as it passes into reservoirs and then through the hydropower electricity turbines.
Think-Outside wished to expand its current data and forecasting offerings so that they are able to provide additional value to its clients in the hydropower industry. The goal of this project was the development of a machine learning pipeline to make predictions for both water inflow into reservoirs/lakes and future electricity prices.
The project outcomes
Due to the different requirements of water inflow and electricity price modeling, the project was divided into two sub-projects: (1) Water inflow prediction and (2) Electricity price prediction. For both sub-projects, we defined the task objectives relating to the project goals specified by Think-Outside, explored numerous data sources, downloaded and processed data (in an automated manner where possible), and then cleaned, explored, and preprocessed the data.
Finally, machine learning models we created and trained on the input datasets and then these models could be used to make predictions about future values of water inflow on a per reservoir/catchment area level and also to predict future values of electricity prices of different energy bidding zones within Norway. The performance of the machine learning models and predictions was analyzed using a number of different error metrics as well as visual representations of the model performances.
A global team of 50 AI engineers collaborated to estimate the total cultivated area and growth stages for fields growing chili for a given district in India states.
The partner for this project, Farm-Hand, is a software and data analytics company (UK and India based) that uses satellite data, and AI/ML alongside initial farmer-led data to obtain field-level insights.
The Indian Farming sector is dominated by the smallholder farmer who on average is managing an area of 2 acres. Many smallholder farmers are affiliated with aggregator companies, whose role may include access to debt finance, provision of seeds/solutions (e.g. fertilizer), farm management practices, and access to the market. The project partner, Farm Hand, is providing its farm management platform to a number of Aggregator Companies who work with chili farmers in the States of Karnataka and Andhra Pradesh in Southern India. Approximately 30 meta-data farm/field attributes are collected, as each farm is onboarded to the platform. Paramount among these with respect to this project are growing season dates for chili production over the last 12 months. In addition, field boundaries are defined and stored as KML files.
Crop growth progression between the dates defined, for the fields identified can be tracked using remotely sensed data from Sentinel Satellite. The Copernicus Open Access Hub provides complete, free and open access to Sentinel-1, Sentinel-2, Sentinel-3, and Sentinel-5P user products, starting from the In-Orbit Commissioning Review (IOCR). Sentinel Data is available via the Copernicus Data and Information Access Services (DIAS) through several platforms. The Institute of Crop Science and Resource Conservation (INRES) at the University of Bonn maintains a Remote Sensed Index Database (IDB). The IDB provides a quick overview of which indices are usable for a specific sensor and a specific topic. In addition to these indices, ESA has developed a number of models, based on ground truth data for eliciting specific features of vegetative growth from remote sense data that include Leaf Area Index (LAI), Fractional Cover (FCOVER), and Fraction of Absorbed Photosynthetically Active Radiation(FAPAR). These indices have been used by many research groups for detecting field boundaries and for creating crop classification algorithms, for instance, Orynbaikyzy et al., 2019, Saini et al., 2018.
Farm Hand provided a dataset containing an initial 300 individual entries with the following information:
Defined field boundaries of a field where chili cultivation has taken place.
The sowing ad harvest dates of at least one chill crop cycle for each field.
Over the duration of the project, additional farms have been onboarded to the platform at an approximate rate of 100 farms per week.
The project outcomes
The objective of this project was to build a machine learning model to detect chili crops and the boundaries of farms containing chilies – for a specific region of India.
The initial task was to gather and understand research that had been conducted for similar projects. Ground truth data, that included farm boundaries and crop cycle information was provided by FarmHand. Satellite imagery was obtained from Sentinel Hub and Google Earth Engine. The model development consisted of two main tasks:
Field Boundary Detection/Delineation Modelling (FBDM), and
Crop Classification using Clustering
For FBDM, the team followed 2 different approaches based mainly on François Waldner’s authored ResUNet-a and FracTAL ResUNet architectures. Both approaches followed Sherry Wang’s paper on field delineation in smallholder farming systems with transfer learning.
For Crop Classification, we used the k-means clustering method where the input was four-time series data sets consisting of the following indices- NDVI, SAVI, NDWI, and EVI (see discussion below) These were calculated from Sentinel-2 images downloaded for the chili crop cultivation cycle (July through March). We used Google Earth Engine to download images (utilizing only images with less than 10% cloud cover) for the cultivation cycle during the year 2019-20. By overlaying the known farm boundaries from FarmHand data, we could improve our confidence in identifying and predicting other farms we believed to contain chili.
Chilli crop clusters with known chili farms layered above (chili = red)
This project tackled the problem of identifying power grid lines by applying GIS and machine learning. The team worked on data collection, preprocessing, modelling, deployment, and maintenance.
The project partner, SurplusMap, is a geo-intelligence SaaS platform that empowers decision-makers in the green transition to make better and faster decisions on where to develop new green industries.
The problem that SurplusMap is trying to solve is to identify and map the power grid from satellite images using machine learning. As seen in the images below, our objective has been to use satellite imagery and other geospatial data in conjunction with machine learning approaches to automatically identify power grid infrastructures such as pylons, substations, and power stations.
Powerstation images at different zoom levels and angles, Source: Omdena; SurplusMap
The use of satellite imagery is important because it can provide an accurate picture of the location of these infrastructures. This allows us to better understand where they are and what proportions of them there are, as well as their size. It also helps us determine whether there are any new power lines being built or if they have been damaged by natural disasters like hurricanes or earthquakes.
The project outcomes
Different object detection pre-trained models were applied. The team also custom-trained and tested various models with new test cases for improvement. The winning models were custom trained with the final data to detect pylons and stations. The model is successfully detecting the objects and will be improved with adding more data. An example visualization of the model can be seen in the image below.
Example visualization using object detection for power infrastructure, Source: Omdena; SurplusMap
India has a variety of geographical features and different climatic conditions in different regions. The project can be scaled up to be applied to other countries for green transition which has geographical features similar to India.
A global team of 50 AI changemakers collaborated in this high-impact 2-month challenge to build solutions for reforestation in Madagascar.
Madagascar-based impact startup Bôndy aims to build innovative reforestation solutions to create social and environmental impacts. They plant high-value species on farmer lands to develop substitutes for the remaining primary forest and to help the rural populations have more sustainable revenues.
Madagascar is the 5th poorest country in the world and faces terrific consequences due to climate change.
The project goals
For Bôndy´s partners, you will help to build a solution that is able to show more transparency.
For their farmer partners, you will help to build an AI model that will allow monitoring more effectively the lands to reforest. Indeed, planting trees is simple but making sure they will grow is more complex. For the moment Bôndy monitors trees physically, but this is not a scalable model. Together with the team, you will help to develop a solution that can combine satellite imagery, transparency, and most importantly that is precise to monitor plants.
This project involves data collection.
Why join? The uniqueness of Omdena AI Challenges
A collaborative experience you never had in your working life! For the next eight weeks, you will not only build AI solutions to make a real-world impact but also go through an entire data science project lifecycle. This covers problem scoping, data collection, and preparation, as well as modeling for deployment.
And the best part is that you will join a global and collaborative team of changemakers. Omdena AI Challenges are not a competition or hackathon but a real-world project that will take your experience of what is possible through collaboration to a new level.