Skip to content

AI Water Wells

Desk studies

The first step is to do some type of desk study. Which in my opinion seems to have the highest ability to automate and improve with AI and LLM’s.

Here are some of the questions given as the types of questions that you might want to answer in a desk study.

  • How much water is required?
  • What water quality criteria apply – potable, livestock, irrigation, other?
  • How many wells are likely to be needed?
  • How much hydrogeological information is already available?
  • What are the data gaps to be filled by additional investigations?
  • What are the social, environmental, community and land ownership criteria that will influence the siting and operation of the well scheme?

Additional questions as generated by chatGPT

  1. Water Requirements:
    • What is the estimated daily and annual water requirement?
    • How might water needs change seasonally or over the long term?
    • How are water needs expected to evolve in response to population growth, development, or changes in land use?
  2. Water Quality:
    • What are the specific water quality standards that need to be met for the intended use? (e.g., drinking water, agricultural irrigation, industrial use)
    • What types of water quality testing will be necessary?
    • Are there any known water quality issues in the region (e.g., presence of natural contaminants like arsenic or fluoride, salinity problems)?
  3. Number and Type of Wells:
    • Based on the water requirement, how many wells will likely be needed?
    • What type of well construction is most suitable given the hydrogeological context (e.g., dug wells, drilled wells, tube wells)?
    • What are the anticipated depths of the wells?
  4. Hydrogeological Information:
    • What is the nature of the aquifers in the region (e.g., unconfined, confined, semi-confined)?
    • What geological formations are present and what are their characteristics (e.g., permeability, porosity)?
    • Are there existing hydrogeological maps, studies, or well logs available?
  5. Data Gaps:
    • What specific data is missing that will need to be gathered through additional studies or investigations?
    • How will these additional studies be conducted (e.g., geophysical surveys, test drilling)?
    • How might data gaps affect the project timeline and budget?
  6. Social, Environmental, Community, and Land Ownership:
    • What are the local cultural, social, or traditional norms or practices related to water use that might influence the project?
    • Are there any specific environmental protection regulations or protected areas that need to be considered?
    • Who are the stakeholders in the community, and how will they be involved or impacted by the project?
    • Who owns the land where the wells will be located, and what permissions are required for access and use?
    • Is there potential for conflict or disputes over water resources or land rights?
  7. Regulatory and Legal Considerations:
    • What local, regional, and national laws and regulations apply to well drilling and water use?
    • Are any permits or licenses required to carry out the project?
    • What are the procedures for well registration or monitoring after the project is completed?
  8. Project Implementation:
    • Who will be responsible for the operation and maintenance of the wells?
    • What resources will be needed for well construction and operation, including human resources, equipment, and materials?
    • What is the estimated timeline and budget for the project?
    • How will the project's success be measured and monitored?
  9. Risk Management:
    • What are the potential risks or challenges associated with the project (e.g., geological risks, technical challenges, social or political risks)?
    • How will these risks be managed or mitigated?
  10. Sustainability Considerations:
    • How can the project ensure the long-term sustainable use of the water resource?
    • Are there measures in place to monitor and manage groundwater levels to prevent over-extraction?
    • How will the project contribute to the broader social, economic, and environmental sustainability of the community?

All of these questions are themselves complex enough that a language model alone might not be able to handle any one of them. However, you can start to understand the project as striving to answer these types of questions.

They would also need some form of answer. What do answers look like. What are the possible ways that answers need to be presented? What sorts of data might we need to answer these questions.

Moving from questions to answers

The questions for the desk study here are only just some of the types of questions that could be asked. But I’m trying to build a system that has enough generality to be able to answer all of the questions, and then give some answer to the end user.

A question like ‘How much water is required’ is very broad and so getting specific enough about this type of question is important. Answers need to be robust enough that you can ask the question and then get back a refined enough answer on the topic.

The questions are supposed to be supported by some data for the area. There is a distinction between the questions straight away. A question like ‘ how much water is required’ is setting up parameters for other questions. If the volume of water is very large, then we might have to ask a broader question about where to find such volume of water.

Where as other questions are more specific and pointed. Are there any electricity lines in the area. The pipeline to answering this question is where our spatial data starts to come in. You would need some map of the areas electricity lines. A layer containing the electicity lines is used and then a search is done over the area of interest. before you might load in this layer, then define some search area and then do a search of the layer with electricity lines. You might have went and done a SQL search for the geospatial. But this can all be mapped to some SQL query my the language model.

Getting data

One of the key things here is being able to query the right datasets. There is a list of datasets that might be used for all of these questions. Not all of the questions can imediatelly be answered with the datasets alone. But they will get you some of the way. Having all of that data in one central place is useful though. There might be topographical, sat data, imagery etc.

One of the interesting things about doing things with AI like this is that you might want to have context data selected and then do your own selection. This is where the context selection stuff comes in to play. More focussed and simple questions might mean that you can just consult one single dataset. But then for a more difficult dataset, you might have to choose the right context. The goal would be to start to ask any type of question and have the models be guided or enquire about projects in ways that could be tested or validated. But for a start you can combine data and get the models to do search over the data.

A first step is being able to access that data easily. You can do this already with CARTO in a way. Where you can select some data from their data catalog and then use that to create some map.

For the case of ireland. It might be a first checkpoint to be able to get all of the data in ireland and then serve it in the correct formats.