AquaSense: Groundwater Level Prediction for Sustainable Agriculture

Overview

AquaSense is an innovative iOS application designed to empower small and medium-sized farmers in France with advanced groundwater level predictions and agricultural insights. Leveraging machine learning and comprehensive data analysis, the app helps farmers make informed decisions about crop selection and water management in the face of increasing climate challenges.

Problem Statement

Climate change is significantly impacting agricultural water resources in France:

• 36-40% of agricultural water needs come from groundwater

• Increasing drought conditions threaten agricultural sustainability

• Farmers lack reliable tools to predict water availability and optimize crop selection

  1. Groundwater Prediction Real-time groundwater level predictions based on: Farmer's specific location Integrated weather API data Machine learning predictive modeling 2. Crop Profit Analysis Comprehensive crop recommendation system Calculates: Hardiness zone assessment Additional water requirements Water cost projections Seed and growing expenses Provides clear, actionable recommendations for most profitable crops

  Technical Architecture

  • Machine Learning Model
  • Weather API Integration
  • Crop Database Management
  • User-Friendly Interface

  

  ---

  Technical Implementation

  We have created a custom API that:

  1. Receives the farmer’s location input
  2. Retrieves weather data and groundwater predictions
  3. Queries the Perenual API for suitable crops based on the hardiness zone
  4. Integrates this information into our analysis model
  5. Returns comprehensive predictions including:
     • Expected groundwater levels
     • Suitable crop recommendations
     • Predicted additional water costs

  

  ---

  Groundwater Prediction Model

  Dataset

  Our groundwater prediction model is trained on a comprehensive French piezometric stations dataset containing:

  • Data from thousands of monitoring stations across France
  • Over 3 million historical measurements from 2020-2023
  • Key measurements include: &nbsp

    - Groundwater level readings
    - Station locations (longitude/latitude)
    - Measurement depths
    - Department and commune information
    - Detailed hydrogeological data

  This robust dataset enables accurate groundwater level predictions, particularly during critical summer months when water management is most crucial.

  Machine Learning Component

  Hackathon-Winning Solution Powered by AutoGluon

  Welcome to the machine learning component of AquaSense, leveraging the winning solution from the Hi!Paris Hackathon Season 5! 🏆 This project demonstrates the power of AutoGluon, a state-of-the-art automated machine learning (AutoML) framework, to deliver exceptional results with minimal effort. Our groundwater prediction model capitalizes on this robust framework to provide highly accurate predictions for sustainable agriculture.

  ---

  🚀 Method Overview: Why AutoGluon?

  Motivation

  Machine learning is often a complex process involving:

  • Model selection
  • Data preprocessing
  • Parameter tuning

  AutoGluon simplifies this by automating the pipeline, enabling rapid development without sacrificing performance.

  Why Use Ensemble Methods?

  Ensemble learning, the foundation of AutoGluon, combines predictions from multiple models to improve accuracy and robustness. This approach is highly effective because:

  • Different models capture different patterns in data.
  • Aggregating multiple models reduces the risk of overfitting and underfitting.
  • Ensembles often outperform individual models in real-world scenarios.

  By automatically training diverse models (e.g., decision trees, gradient boosting machines, neural networks) and combining their outputs, AutoGluon delivers robust and reliable results.

  ---

  State-of-the-Art Features of AutoGluon

  • Automatic Ensembling: Combines models using bagging and stacking without requiring manual intervention.
  • Out-of-the-Box Performance: Delivers strong results without hyperparameter tuning, making it ideal for tight deadlines.
  • Robust Handling of Real-World Data: Seamlessly processes missing values, categorical variables, and imbalanced datasets.
  • Resource-Aware: Adjusts to available computational resources for optimal efficiency.

  ---

  Why AutoGluon for AquaSense?

  AutoGluon’s simplicity and effectiveness made it the perfect choice for AquaSense. We leveraged its default settings, relying entirely on its:

  • Predefined ensemble methods
  • Automatic preprocessing capabilities
  • Baseline hyperparameters (no tuning required)

  This approach allowed us to focus on understanding the problem and the data, rather than fine-tuning the model.

  ---

  📈 Results

  Key Metrics for Hi!Paris Hackathon Season 5:

  • F1 Score: 69.61%

  AutoGluon delivered superior results by:

  • Automatically ensembling multiple models
  • Handling data preprocessing efficiently
  • Using robust baseline configurations designed for competitive performance

  ---

  Crop Data Integration

  Crop Data API

  We utilize the Perenual API (perenual.com), which provides comprehensive plant data, including:

  • Watering needs
  • Hardiness zone compatibility
  • Growth requirements
  • Seasonal information

  This data helps us determine crop suitability and water requirements for different locations.

  ---

  Reports

  1. Business Report
  2. Scientific Report

  ---

  📜 License

  This project is licensed under the MIT License.

  ---

  Contact

  Developed as an open-source innovative solution to address agricultural water management challenges in the context of climate change.