Obsidian/Doktoritöö/000_otsast_peale/00.02 - kontekstid/Overall Goal and Core Problem.md

Indirect‑methodology for estimating traffic volume on rural local roads

Below is a step‑by‑step framework that builds on approaches documented in the literature and can be adapted to your study area.

Step	What to do	Why it helps / supporting evidence
1. Define the target network	Compile an inventory of all local‑road segments for which you need traffic volumes (length, functional class, geometry, surface type).	A clear road‐network database is a prerequisite for any modelling effort.
2. Gather surrogate variables	For each segment collect readily available attributes that correlate with traffic demand: • Demographic data – household count, population density, employment numbers (often available from census or municipal GIS). • Land‑use / land‑cover – proportion of residential, agricultural, commercial areas. • Road geometry – lane width, shoulder width, curvature, intersection density. • Connectivity metrics – distance to major arterials, presence of state‑road intersections.	Studies have shown a strong linear relationship between total entrance AADT and the number of households in a community; this can be leveraged for rural networks [5].
3. Select an estimation technique	Choose one (or combine several) of the following proven methods: • USF/URS regression models – use traffic‑count data from counties with good coverage to calibrate a statistical relationship between AADT and the surrogate variables, then apply it to un‑monitored segments [2]. • Proposed hybrid method (as described in source 2) that blends USF/URS outputs with ground‑truth counts for validation.	The USF and URS methods have been evaluated against observed AADT and performed well when sufficient count data exist; the proposed hybrid approach further improves robustness where data are sparse [2].
4. Develop a calibration dataset	Identify a subset of local roads that do have recent traffic counts (e.g., from permanent counters, short‑term manual surveys). Use these as the dependent variable in your regression model.	A calibrated model ensures that relationships derived elsewhere are adjusted to local conditions.
5. Fit the statistical model	Apply a suitable regression technique (ordinary least squares, negative binomial, or zero‑inflated models) using the surrogate variables as predictors of AADT. Test alternative specifications and retain those with significant coefficients and good goodness‑of‑fit statistics.	Negative‑binomial approaches have been successfully used for accident prediction and can handle over‑dispersed traffic count data [3].
6. Apply DVMT adjustment factors (if needed)	After obtaining preliminary AADT estimates, adjust them using DVMT‑based correction factors that account for differences between modeled vehicle‑miles‑travelled and observed county‑level totals. For rural segments the factor is simply the ratio of KYTC‑derived DVMT to model‑estimated DVMT; for urban‑type local roads a two‑scenario adjustment (increase for state‑road intersections, decrease otherwise) can be used [4].	This step prevents systematic over‑ or under‑estimation on roads that either feed into higher‑order networks or are isolated.
7. Validate the results	Use empirical Bayes techniques to compare predicted AADT with any independent counts not used in calibration, quantifying percent reduction in error (e.g., a 10–15 % improvement is typical).	Empirical Bayes has been employed to assess safety‑treatment effects and can be repurposed for volume validation.
8. Produce the final traffic‑volume database	Populate each unmonitored segment with its estimated AADT (and optionally VMT = AADT × segment length). Document confidence intervals derived from model standard errors or bootstrap resampling.	Providing uncertainty bounds is essential for downstream safety or planning analyses.

Practical tips & low‑cost data sources

1. Remote sensing / satellite imagery – Recent high‑resolution satellites can classify vehicle presence and, when combined with road geometry, give a rough traffic‑density proxy (useful as an additional predictor).
   
2. Mobile‑phone based positioning – Even coarse multilateration (~50 m accuracy) yields aggregate travel‐flow patterns that can be calibrated to AADT for larger corridors; however, resolution may be insufficient for short rural links.
   
3. Community‑level models – If household data are the most reliable surrogate in your region, you can directly apply the linear relationship from source 5 (AADT ≈ k × households) after calibrating coefficient k with local counts.
   

Example workflow illustration

1. GIS → road inventory + census blocks → household count per segment  
2. Compile predictor table: {households, %residential land‑use, lane width, shoulder width, curvature}  
3. Calibrate USF regression on 30 counted segments (R² ≈ 0.72)  
4. Apply model to 200 uncounted segments → preliminary AADT values  
5. Compute rural DVMT factor = (County DVMT from KYTC) / (Sum of modeled VMT) → adjust each segment’s AADT  
6. Validate on 10 hold‑out counts → Empirical Bayes reduction in error ≈ 13%  
7. Export final CSV with AADT ± 95 % CI for all segments

Summary

• Start with easily obtainable surrogate variables (households, land use, geometry).
  
• Calibrate a regression model using the USF/URS framework or the hybrid method described in source 2.
  
• Refine estimates with DVMT‑based adjustment factors as per source 4 to align network‐wide vehicle miles with observed totals.
  
• Validate through empirical Bayes or independent counts, and provide uncertainty metrics.
  

Following this structured approach will give you a defensible, cost‑effective estimate of traffic volumes on rural local roads without the need for extensive field counting campaigns.