All 8 Processing algorithms in the PlanX DataCube Lab provider. Each entry includes the algorithm ID (for PyQGIS), its purpose, and a link to the online documentation.
planx_datacube:create_data_cube
Construct a multi-dimensional (X×Y×T×V) data cube from vector or raster time-series layers. For vector input: select multiple layers, each representing one time step. The algorithm rasterises each layer onto a common fishnet grid and builds the cube. For raster input: select multiple single-band rasters. Output: netCDF4 file and/or in-memory cube for downstream EHSA / forecasting.
planx_datacube:space_time_cube
Aggregate point/feature events into a regular space-time cube: a fishnet grid crossed with time bins. Each feature is binned by its centroid into a grid cell and by its time field into a time step, then aggregated (count / sum / mean / max / min). The output is a long-format panel polygon layer that feeds directly into EHSA, Local Moran's I, Trend, Anomaly and Forecast tools. Optionally writes the cube to netCDF4 for voxel rendering.
planx_datacube:emerging_hot_spot_analysis
Perform Emerging Hot Spot Analysis on panel data. Classifies each spatial unit into one of 17 patterns using Getis-Ord Gi* z-scores and the Mann-Kendall trend test. Spatial relationship: fixed distance band, K nearest neighbours (recommended), or global standardised z-score. Output: one row per location with ehsa_code (0-16), ehsa_label, mk_tau, mk_p_value, mk_slope.
planx_datacube:local_morans_i
Anselin Local Moran's I (LISA): find statistically significant spatial clusters and outliers. The panel is first reduced to one value per location (mean, last, sum or max over time), then each location is classified as High-High / Low-Low cluster, High-Low / Low-High outlier, or Not significant. Significance via deterministic conditional permutation test.
planx_datacube:temporal_trend
Detect monotonic trends per location with the non-parametric Mann-Kendall test and the robust Theil-Sen slope estimator. Each location is classified as Increasing, Decreasing, or No significant trend at the chosen significance level. Output: location_id, n_obs, mk_tau, mk_p_value, sen_slope, sen_intercept, trend_code, trend.
planx_datacube:temporal_anomaly
Flag anomalous time steps per location using a robust modified z-score (median / MAD) — resistant to the very outliers it looks for — or the classic mean / standard-deviation z-score. A point is anomalous when |z| >= the threshold (default 3.0). Output: location_id, n_obs, n_anomalies, anomaly_rate, max_abs_z, last_z, last_anomaly, first_anom_time, last_anom_time.
planx_datacube:forecast_backtest
Backtest the forecasting methods per location: hold out the last N observations, forecast them from the earlier history, and score each method by MAE, RMSE and MAPE. The method with the lowest hold-out RMSE is reported as best_method — use it to decide which method to trust before forecasting forward with the Time-Series Forecast tool. Output: location_id, n_obs, best_method, best_rmse, and per-method RMSE columns.
planx_datacube:time_series_forecast
Forecast future values per location using the selected method. Auto mode tries ARIMA first, then Random Forest, then Exponential Smoothing. Output: one row per location with forecast columns (forecast_t1..tN) and the method actually used (forecast_method).