GeoAI in resource-constrained environments

Marc Böhlen,Gede Sughiarta,Atiek Kurnianingsih,Srikar Reddy Gopaladinne,Sujay Shrivastava,Hemanth Kumar Reddy Gorla

2024-08-30

Abstract:This paper describes spatially aware Artificial Intelligence, GeoAI, tailored for small organizations such as NGOs in resource constrained contexts where access to large datasets, expensive compute infrastructure and AI expertise may be restricted. We furthermore consider future scenarios in which resource-intensive, large geospatial models may homogenize the representation of complex landscapes, and suggest strategies to prepare for this condition.

Computers and Society

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: how can small - scale organizations (such as non - governmental organizations, NGOs) effectively utilize Geo - Spatial Artificial Intelligence (GeoAI) technology in a resource - constrained environment. Specifically, the paper explores how to develop and apply GeoAI methods suitable for these organizations to support their needs for managing and analyzing land resources in the case of limited data, computing resources and expertise.

### Core issues of the paper
1. **Resource limitations**:
- Small - scale organizations usually lack large - scale data sets, expensive computing infrastructure and AI expertise.
- These limitations make it difficult to apply complex and resource - intensive large - scale geospatial models in a resource - constrained environment.

2. **Geographical coverage imbalance**:
- Existing large - scale GeoAI models are mainly trained for land cover categories in the Northern Hemisphere, ignoring the special needs of other regions such as tropical rainforests and deserts.
- Western - centered land management analysis may lead to misunderstandings or misinterpretations of the complex landscapes in developing countries.

3. **Combination of local knowledge and technology**:
- How to combine local knowledge with advanced GeoAI technology to better serve the needs of specific regions.
- For example, through a case study in central Bali, Indonesia, the paper shows how to help the Wisnu Foundation use GeoAI tools to support its community resource management and land mapping work.

### Solutions and methods
The paper proposes several different GeoAI approaches aimed at balancing resource limitations and technological complexity:

1. **Local institutions, small data sets, low computational complexity**:
- Use simple machine - learning algorithms (such as Random Forest, RF), implemented through the QGIS framework, suitable for low - performance hardware devices.
- The advantage is that it is easy to start with, and the disadvantage is that its scalability and flexibility are limited.

2. **Local institutions, small data sets, relatively high computational complexity**:
- Use Support Vector Machines (SVM) combined with a cloud - computing environment, which improves accuracy and ease of use.
- Provide a customized cloud environment, Cocktail, for easy remote - team collaboration and result sharing.

3. **Computational flexibility, extensible models, large data sets**:
- Use the U - Net model, which can adapt to different hardware configurations from CPU to GPU.
- Support larger - scale data and models, but at a higher cost, suitable for long - term investment.

4. **Standard data sets and pre - trained models**:
- Combine large - scale general - purpose data sets and large - scale neural - network models, such as the Functional Map of the World (FMW).
- Aim to create machine - learning models capable of predicting functional land use, but need to consider the cost and applicability of data - set acquisition.

### Summary
The core objective of the paper is to explore and develop GeoAI solutions suitable for small - scale organizations in a resource - constrained environment, ensuring that these organizations can fully utilize geospatial artificial intelligence technology and enhance their capabilities in land management and environmental protection.