Research on current methods relating to spatial aggregation of Renewable potential

[GbotemiB]

what is the current status on the research related to spatial aggregation of the renewable potential. Here are two related topics:

• an options to have multi-solar and wind power profile per zone multi-solar and wind power profile per zone pypsa-earth#699 [the itself discussion could have been outdated but it gives a good starting point]
• Martha's PhD study on the effects of spatial clustering on power modelling outputs which main results are published in this paper https://link.springer.com/article/10.1186/s42162-022-00187-7. There is a solid computational work behind but it sin't really conclusive in what relates any practical guidance on choosing of good enough spatial resolution or compatibility of the defended clustering algorithm with grid representation

[GbotemiB]

Overview

Following the investigation into methods for harmonizing generation and transmission data, two key areas were identified to improve the accuracy of spatial aggregation in the modeling workflow:

Handling Heterogeneous Renewable Profiles (Wind/Solar)

The Problem: Current spatial clustering based solely on distance often aggregates distinct geographical features together (e.g., a mountain peak with high wind intensity and a valley with low intensity). This leads to profiles "canceling out" or averaging down, obscuring high-potential sites.

Proposed Solution: Implement Class-based Aggregation (referencing Max's suggestions). Instead of aggregating purely by distance, wind/solar profiles should be segregated into classes. This ensures that high-quality resource sites are preserved in the model rather than being smoothed out by lower-quality neighboring areas.

Mismatch between Renewable Potential and Grid Capacity

Issue: Distance-based clustering often snaps high-potential generation sites (e.g., Amazonia solar) to nearby substations with insufficient capacity.

Impact: This leads to artificial bottlenecks and high curtailment in the model, as the physical grid cannot evacuate the theoretical power generated. The clustering must account for the capacity of the connection point, not just the distance.

Network Topology and Clustering Algorithms

The Problem: Standard K-means clustering ignores physical network constraints. In complex terrains (e.g., "Andean topology"), two nodes might be geographically close but electrically distant (not connected directly due to terrain, requiring a long loop). K-means may incorrectly cluster these together.

Proposed Solution: Use Hierarchical Agglomerative Clustering (HAC) (referencing methods from Martha's paper). This method respects the structure of the network and avoids creating unrealistic connections between nodes.

Optimization Strategy: Raw HAC is computationally expensive. To mitigate this, we can utilize pre-connected node logic (based on OSM data/polygons). By only clustering nodes that are already physically connected in the raw data, we reduce the search space. This acts as a "constrained" HAC, offering a balance between topological realism and computational efficiency.