H-GAC’s Forecast as “Production” Operations: General Organization, Logistics, and Schedules Dmitry Messen Houston-Galveston Area Council
Overview Forecasts: 1972, 1986, 1992, 2003, 2006 Breaking up with the RTP cycle Beginning in April 2013: Quarterly Releases – Annual data – Population, Employment: TAZ, CT, County, Region – Land use (type, sqft, HU): parcel Current, Announced Changes, Model Predictions No “adoption” process, no drafts/preliminaries (de facto Public) continuous external/internal Review focused on detection of factual “errors” (QA/QC)
2-Phase Forecasting System Demographic Microsimulation – Population regional “totals” – Net change in households (demand for housing) – Labor force Workforce Jobs (demand for non-res sqft) Land Use Microsimulation – Aggregate demand List of Projects – Development Proposals Selection on ROI Buildings 3 rd Phase (not yet implemented) – Household and job location choice
LU Data Development Continuous, never ending process Independent from model development and model execution Labor-intensive and code-intensive Requires smart and flexible design of workflows and data architecture No way around it
Dynamic Land Use Corrections to Current LU (existing buildings) – Why live with errors? Why not keep corrections? Announced Projects – Regionally-significant (e.g., new Exxon’s campus) – Locally-significant Sources – Imagery, Google’s StreetView – Plats (ordered from counties 4 times a year) – Business media – Appraisal (once a year)
Parcels and Buildings Create and maintain our own polygons (“Master Polygons”) and buildings From land ownership parcels to polygons that are – Positionally accurate – Comprehensive (100% coverage) – Integrated (land, water, roads) – Meaningful (single polygon for a park, downtown block, mall, etc) Detect annual changes in parcels and apply them to “Master Polygons”
Parcels and Buildings Keep dynamic info (valuations, rents) separate from static info (type, sqft, floors) Tie-backs to appraisal records Land Use data (GIS: feature classes and tables) Create from Land Use data (+ other data) – Inputs for model simulation – Inputs for model estimation (we do not re- estimate the model every quarter)
Non-LU Inputs ACS Summary Tables, ACS PUMS, BEA, BLS Once a year, but schedules are different There’s always some update that we can include in a quarterly release Efficient process, takes minimal time – download, run SAS code Supports Currentness – Latest Planning Assumptions (Fed regs)
Decoupling of Production and R&D Production – Product (Forecast) is always available – Updated/Upgraded quarterly – Releases are labeled (2014Q1, 2014Q2,…) R&D – Model changes – New components
Distribution and Review Distribution – Table query tool; download xls – Web-based mapping app (RLUIS); download GIS data – Map service Review – Parcel-specific feedback directly from RLUIS (also TAZ and CT) – General comments
Staffing (8) Manager Senior Modeler (LU) Senior GIS Analyst (Data Development) – 2 GIS Analysts and 1 GIS Technician Senior Analyst (Tools, Technology, Infrastructure) – 1 GIS Analyst/Programmer
Lessons Credibility of our work hinges on the accuracy of the current land use and development “pipeline” data Enormous potential for other applications – Community planning, public health, environmental Benefits of transparency and openness Challenges (difficult=interesting) – Design of workflows and procedures – Technology (SDE, network, web) – Distractions; escalating expectations
?