Short‑term forecasting of cloud‑induced power ramps across a network of distributed photovoltaic (PV) or irradiance sensors is a well‑known pain point for grid operators. A natural idea is to make the graph neural network (GNN) advection‑aware: connect each site to its upwind neighbours and set edge time‑lags according to the cloud‑motion vector (CMV), so that a ramp is propagated before it physically arrives. Using a synthetic testbed with a known wind field, we obtain three key findings:
(i) With a realistic cross‑correlation CMV estimate, an explicit advection graph does not outperform a plain static or learned‑adjacency spatiotemporal GNN;
(ii) Roughly half of the benefit of a perfect CMV comes simply from providing the accurate motion vector as an input feature, not from the graph structure itself;
(iii) Advection helps only when the advective displacement over the forecast horizon $v*H$ fits inside the sensor network.
Motivated by (ii), we introduce a compact self‑supervised cloud‑motion estimator – a position‑aware encoder trained solely on a multi‑lag optical‑flow reconstruction loss with an annealed kernel. The estimator recovers the true wind vector with a median angular error of 2‑4°, i.e., 2‑4× better than the classical cross‑correlation method across all wind regimes. Freezing this estimator and feeding its vector to the forecaster closes about 60% of the oracle‑CMV RMSE gap at moderate wind (8‑15% RMSE reduction over no advection), without any external wind data.
We also report a negative result for a spatially coherent probabilistic head. All claims are established on a single synthetic simulator; we discuss why validation on real networks is the necessary next step and outline a possible approach.
Review: The paper clarifies the limited value of advection‑aware graph structures and demonstrates that a self‑supervised motion estimator can capture most of the gains, offering a practical route for improving short‑term solar ramp forecasts in real deployments.