Predict surface reflectivity for roofs using high resolution imagery

**Input Data:** - geometries of features of interest (building footprints for roofs). Official building footprints from LA city government was used for LA and Microsoft footprint data was used to acquire roof geometries for other cities. - high-resolution four-band imagery for features of interest for times of interest. NAIP was used as the primary source of imagery but the model was also tested on Airbus Pleiades imagery. The initial results on Airbus imagery was also very promising. The input data consist of 100 rows and each represent data for individual roofs in LA county. This is a sample of 100 roofs from our master dataset of all the roofs in LA county. Each row contain 16 columns. "norm_red_mean", "norm_green_mean", "norm_blue_mean" and "norm_nir_mean" are the main inputs and represent preprocessed, normalized mean band values calculated from all the pixels within a roof for the year 2009. NAIP was used as the source of imagery. **Model description:** A decision forest regression model trained in this [experiment][1] was used to make prediction. **Output:** The "Scored Label Mean" column from the "Score Model" module represent the albedo prediction for each roof in 2009. This output will be used to create the final product which is a vector dataset with estimated mean reflectivity (unitless albedo, 0-1) for each feature of interest (roof, street segment, etc.) for each time of interest. Possibility to also produce a raster version of this dataset--estimating the reflectivity at each pixel. [1]: https://gallery.cortanaintelligence.com/Experiment/Train-a-decision-forest-model-to-predict-surface-reflectivity-of-roofs