Publication date

Apr 20, 2020

Edition

2020-04-01

Abstract

Exposure Map for Tanzania, by ImageCat, 2020-04-01.  

Click on the map to view a distribution of costs by construction type.

This data was collected for Tanzania as part of the Modelling Exposure Through Earth Observation Routines (METEOR) project and is a Level 3, regionally enhanced data set. The purpose of collecting the data is CAT modeling and loss estimation. Modeling potential exposure helps mitigation and response planning by identifying regions of high risk or greater exposure before a natural disaster occurs. Repurposing for any reason other than assessing risk is not recommended. The data presents the estimated number of buildings, building area, and rebuilding value at a 3-arcsecond grid resolution for Tanzania (approximately 90 meters, depending on where on earth). The results were created through a process of spreading the number of buildings from census data, at the district census level, to the 3 arcsecond level by a statistical assessment of moderate resolution EO data, which is described in more detail below. The estimated building count at any given area is a result of statistical processes and should not be mistaken as a building count. The structural classes of buildings used for risk assessment are estimated given the building wall, floor, and roof material classes surveyed through the IPUMS data set. Structural classes for urbanized areas such as Dar es Salaam were developed through field observations using stratified sampling, an engineer, and local expert verification.

Data Quality - Lineage

Dasymetric Mapping Evaluation:

A visual review and adjustment by 3 different analysts not involved in the process- junior level, mid-level, and senior level was conducted. Comparison with several dasymetrically mapped products, including Landscan, GHSL-SAR, GUF, and WorldPop were reviewed in Google Earth with particular attention to low population areas.

Structural Distribution Evaluation:

Structural Distribution was developed through the implementation of the stratified sampling method by Humanitarian OpenStreetMap Team (HOTOSM) and reviewed by and ImageCat. In addition, the structural mapping from the IPUMS data was validated through mapping the structural classes and review of the results with ImageCat and the HOTOSM team. The distribution of structural classes closely reflected known patterns of construction practices as verified by the in-country structural engineer.

Persons per household Evaluation:

Given detailed census level data available, no additional validation was performed.

Development Pattern Evaluation:

The development patterns were reviewed by 5 different parties and included review of several land use maps on the internet. Concerted attention was given to regions with mixed construction and the mapping of structural types.

Building Height and Area Evaluation:

Building height and size were generated using highly detailed micro census data from HOTOSM building height samples, and OSM building footprints from selected regions. The vast number of buildings are single story. HOTOSM building footprints were selected where the data was complete, and these values were used to develop average sizes. For some samples, these averages were quite small and were excluded as outliers. The resulting data was reviewed spatially on maps by ImageCat.

Replacement costs Evaluation:

The replacement costs were obtained from the GFDRR Africa Disaster Risk Financing – Result Area 5 Exposure Development: Replacement Cost Refinements to the Exposure data report prepared by ImageCat for the World Bank. Significant literature review, local engineers’ input, construction manual or developer’s estimates, and hazard modeling went into developing the replacement cost values, by wall material durability and development pattern, in the GFDRR report. Thus, the replacement cost values were used directly with no additional corrective measures taken.

 

Processing Step Descriptions:

Population Data processing

The ward-level administrative units population information was used in Tanzania, approximately an administrative unit level-3, to determine population and number of households. The ward GIS data was provided by John Mathias Kiriwai from the Tanzanian Prime Minister’s Office (PMO). Along with the GIS shapefiles PMO provided detailed excel spread sheets of the population counts per ward. The persons per household figures were available at the ward data in the PDF and matched to the ward level by converting to a word file and matching manually to adjust for variations in spelling and punctuation.  All values are from the 2012 National Bureau of Statistics housing and population census. The ward-level population and household values were mapped to the IPUMS district-level data set then interpolated to a 3-arcsecond (~100m) grid cell and used to infer the number of buildings.

 

Dasymetric Mapping

Dasymetric mapping is the process of spreading the number of buildings from census data, at the Tanzanian District and Ward census unit level to the 3-arcsecond level, by a statistical assessment of moderate resolution Earth Observation (EO) data. To collect EO indicators of settlements and the density of buildings, various remote sensing data sets were used. These included:

  1. NOAA night-time light annual composite (VIIRS) resampled from the 15-arcsecond grid cell to 3-arcsecond
  2. Oak Ridge National Laboratory Landscan ambient population (LSCAN) resampled from the 15-arcsecond grid cell to 3-arcsecond
  3. JRC Global Human Settlement Layer (GHSL-Landsat) derived from Landsat imagery resampled from the 15-arcsecond grid cell to 3-arcsecond
  4. DLR Global Urban Footprint (GUF) resampled from 12-meter grid cells to a 3-arcsecond percentage of human presence raster grid
  5. JRC GHSL derived from Sentinel-1 SAR (GHSL-SAR) resampled from 20-meter grid cells to a 3-arcsecond percentage of human presence raster grid
  6. CIESIN-Facebook High Resolution Settlement Layer (HRSL) resampled from 1-arcsecond grid cells to a 3-arcsecond percentage of human presence raster grid
  7.  JRC mosaiced Sentinel-1 dual polarization bands (SAR B1, B2, B3) resampled from 20-meter resolution to a 3-arcsecond mean raster grid per band and a maximum mean value of the 3 bands (SAR-MaxMean).
  8. Tanzania gridded population from WorldPop 2020 at 3-arcsecond grid cells
  9. An indicator of OSM data throughout the country- building count, area, and maximum building height were calculated from building footprint polygons and aggregated up to create 3-arcsecond grids.

These remotely sensed earth observation products and building footprint aggregates establish the distribution statistics for dispersing buildings by urban density and development pattern type. Each of the EO products, individually or in a combination, act as weights to disperse the known population per Tanzanian district census level by structural type and determined development pattern to 3-arcsecond grid cells. For example, in grids designated as a development patterns 1 or 2 (resembling rural or single-family residential communities) an even building distribution of the district is reallocated only to grid cells within the district associated to human settlement. For determining the weights for distribution within a given district, several machine learning algorithms were run using the EO to develop a prediction model.

 

Masking process: To prevent unpopulated areas from being considered as settlements a mask was created by combining the extents of night-time light (VIIRS), ambient population (LSCAN), and GHSL-Landsat that have been reclassified to inhabited vs. uninhabited using a minimum threshold value determined by visual inspection. These minimum values correspond to even the sparsest human presence. This mask was used to subset the high-resolution Sentinel-1 mosaic with dual polarization (Band1 = VH, Band2= linear comparison of VH/VV, Band3 = VV) that goes into the machine learning process to come up with the development patterns. The Global Urban Footprint (GUF), Sentinel-1 SAR based GHSL product, and CIESIN-Facebook High Resolution Settlement Layer (HRSL) were not subset because those data sets were good indicators for locating rural settlements.  

Development Pattern Creation: Development patterns are patterns of construction in a given country that typify the building structure development and density as much as possible. They sometimes correspond with land use, but not always. The development patterns are determined by a structural engineer working with GIS analysts to conduct a web reconnaissance survey of available data regarding the structural type and distribution of the country. Various engineering websites and standards (Prompt Assessment of Global Earthquakes for Response [PAGER], Global Earthquake Model [GEM]) were used to establish a preliminary structure type distribution. These preliminary distributions are validated through Google StreetView survey. After the web reconnaissance the structural engineer begins to formulate the development pattern.

To characterize building density in more populated areas, analysts digitized sample training development patterns polygons in the top 10 most populated cities in the country. The basemap vintage and source used during digitization vary by region and zoom level. However, the most current high-resolution satellite images are used. The training polygons and the moderate resolution EO products described above are used in a machine learning process (CART algorithm, Random forest, and Support Vector) for assigning the development patterns throughout the country, which informs the estimated building density. The intensity of urbanity correlates to both the building density and the structural distribution (see Structural Distribution Processing Step for more details). For Tanzania, the ImageCat engineer characterized 6 development pattern types (see below Development Pattern Descriptions).

 

Number of Buildings

The total number of buildings per structural class per grid cell was inferred using a combination of development patterns samples, ImageCat building count sample, and aggregated OSM building count raster. For each development pattern, 3-arcsecond grids were selected that had at least 90% of the buildings digitized in the OSM data. These grids were selected throughout the country, in a variety of urban environments. These were used to train a regression model using the 3-arc-second GHSL-SAR and the GUF. GHSL-SAR proved more effective in highly developed regions, whereas GUF was more effective in rural areas. The model was are constrained using the 3-arcsecond OSM building count raster data sets.

 

Structural Distribution

There are two methods for developing the structural distribution:

  1. Using the IPUMS district census data mapping schemes were developed using a combination of the building wall, roof, and floor material type.  Each combination of wall, roof, and floor material type are mapped to a GEM construction type, and the number of households is used to develop a structural distribution by census district. Where IPUMS district information was not provided, the neighboring district was used as a proxy.
  2. Field survey data collected by HOTOSM was used to establish the mapping scheme for each development pattern. The HOTOSM sampling strategy followed the stratified sampling and Bayesian updating approach suggested in: Porter K., Z. Hu, C. Huyck and J. Bevington (2014), User guide: Field sampling strategies for estimating building inventories, GEM Technical Report 2014-02 V1.0.0, 42 pp., GEM Foundation, Pavia, Italy, doi: 10.13117/GEM.DATA-CAPTURE.TR2014.02. The HOTOSM building survey data was verified by an in-house engineer (Michael Eguchi). All the mapping schemes are then mapped to the PAGER standard structural types. These structure types are overlaid with the manually delineated development pattern sample polygons to create a refined mapping scheme. A final round of sanity checking is conducted by ImageCat engineers.

 

Based on a review by HOTOSM and ImageCat, it was determined that the IPUMS data captured the diversity of construction practices throughout the country and that the HOTOSM captured the diversity in highly urban areas. Thus, the IPUMS mapping schemes were applied to Development Pattern 1, and the HOTOSM mapping schemes were applied in development patterns: 2,4,5,6,7. The exception was Zanzibar, where due to the unique history of the region and homogeneity of the construction, the IPUMS mapping schemes were used for the entire island. 

The data below provides a description of the method of inferring structural class from the census data, used in non-urban areas, for each GEM taxonomy classification.  

 

CONTYPE - Description

CR/LFINF+DNO/HBET:1,3 - Non-ductile reinforced concrete infilled frame (1-3 stories)

CR/LFINF+DNO/HBET:4,7 - Non-ductile reinforced concrete infilled frame (4-7 stories)

CR/LFINF+DNO/HBET:8,20 - Non-ductile reinforced concrete infilled frame (8-20 stories)

CR/LFM/HBET:1,3 - Reinforced concrete moment frame (1-3 stories)

CR/LFM/HBET:4,7 - Reinforced concrete moment frame (4-7 stories)

CR/LFM/HBET:8,20 - Reinforced concrete moment frame (8-20 stories)

MATO/LN - Mixed construction types

MUR+ADO/HBET:1,3 - Unreinforced adobe masonry (1-3 stories)

MUR+CB99/HBET:1,3 - Unreinforced concrete block masonry (1-3 stories)

MUR+CB99/HBET:4,7 - Unreinforced concrete block masonry (4-7 stories)

MUR+CL99 - Unreinforced fired clay masonry

MUR+STRUB - Unreinforced rubble stone masonry

S - Steel

W - Wood

W+WWD - Wattle and Daub (Walls with bamboo/light timber log/reed mesh and post)

 

The mapping schemes below provide the distribution of building by structural type per development pattern. The structure type key can be found below.

Development Pattern 1: Rural development found outside of city boundaries and is typically associated with agricultural development. The regions typically consist of small, remote villages with single roads in and out. Buildings are typically spaced far apart and are almost exclusively 1 to 2 stories. Local materials and construction practices are generally used and performed in these areas. MATO/LN - 0.0233100233100233 | MUR+STRUB - 0.00233100233100233 | W - 0.00233100233100233 | W+WWD - 0.0536130536130536 | MUR+CB99/HBET:1,3 - 0.916083916083916 | CR/LFM/HBET:1,3 - 0.00233100233100233

Development Pattern 2: This development pattern reflects areas typically dominated by single family residential structures. Commercial properties, such as local markets, are present, however residential structures are the primary occupancy. The built-up area is denser than rural class 1, however open land (yards, vacant lots, etc.) are present and can be observed via satellite imagery. All structures are low-rise, with most in the 1 to 2 story range. MATO/LN - 0.00421940928270042 | W - 0.00421940928270042 | MUR+CL99 - 0.156118143459916 | MUR+CB99/HBET:1,3 - 0.827004219409283 | CR/LFM/HBET:1,3 - 0.00421940928270042 | CR/LFINF+DNO/HBET:1,3 - 0.00421940928270042 

 

Development Pattern 4: This development pattern is typically associated with extremely dense settlements using mixed materials. They are usually found within boundaries of large cities and are typically comprised of very small (<100 m^2) standalone structures with little to no space between adjacent buildings. The settlement is unplanned, therefore there is no organization to the configuration of building layouts. Almost all structures are 1-story and are typically erected using cheap and accessible local materials. MATO/LN - 0.036144578313253 | MUR+CB99/HBET:1,3 - 0.959839357429719 | CR/LFINF+DNO/HBET:1,3 - 0.00401606425702811

Development Pattern 5: Development pattern 5 is characterized by urban areas predominantly occupied by low to mid-rise residential and commercial structures. An occasional high-rise apartment or office building may be present. These developments are typically found near or around major city centers. Buildings are tightly spaced and are fairly regular in shape. MATO/LN - 0.0170132325141777 | MUR+CB99/HBET:1,3 - 0.521739130434783 | MUR+CB99/HBET:4,7 - 0.0113421550094518 | S - 0.0056710775047259 | CR/LFM/HBET:1,3 - 0.0264650283553875 | CR/LFM/HBET:4,7 - 0.00756143667296786 | CR/LFM/HBET:8,20 - 0.0132325141776938 | CR/LFINF+DNO/HBET:1,3 - 0.262759924385633 | CR/LFINF+DNO/HBET:4,7 - 0.102079395085066 | CR/LFINF+DNO/HBET:8,20 - 0.0321361058601134

Development Pattern 6: This development pattern is the central business district of urban areas within the major cities. The region is occupied by low to high-rise apartments and commercial offices. Most structures are under 7-stories, however high-rise (8+ stories) can be found within the region. Building footprints are larger than most non-industrial development patterns. This development pattern will be found only in major cities and along the major, paved roads. MATO/LN - 0.0305595765524704 | MUR+CB99/HBET:1,3 - 0.638344226579521 | MUR+CB99/HBET:4,7 - 0.00435729847494553 | S - 0.0152505446623094 | CR/LFM/HBET:1,3 - 0.0435729847494553 | CR/LFM/HBET:4,7 - 0.0196078431372549 | CR/LFM/HBET:8,20 - 0.0479302832244009 | CR/LFINF+DNO/HBET:1,3 - 0.130718954248366 | CR/LFINF+DNO/HBET:4,7 - 0.0457516339869281 | CR/LFINF+DNO/HBET:8,20 - 0.0239651416122004

Development Pattern 7: This development pattern is characterized by areas dominated by ports, mining or industrial activities. Structures are typically closely spaced and regular in shape. A majority of buildings within these regions are warehouses, rectangular shape and single story. Smaller low-rise, office and commercial structures can also be found on site. MATO/LN - 0.00719424460431655 |  W - 0.00479616306954436 | W+WWD - 0.00239808153477218 | MUR+CB99/HBET:1,3 - 0.6810551558753 | S - 0.119904076738609 | CR/LFM/HBET:1,3 - 0.0383693045563549 | CR/LFM/HBET:4,7 - 0.00719424460431655 | CR/LFINF+DNO/HBET:1,3 - 0.127098321342926 | CR/LFINF+DNO/HBET:4,7 - 0.0119904076738609

 

Building Height:

The building height values for all development patterns were obtained from the field survey data collected by HOTOSM. This established the building height by structure type per development pattern type. The HOTOSM survey was reviewed and validated by an in-house engineer using images provided of the individual buildings.

 

 

Total Building Area:

The total building was calculated using a combination of Humanitarian OpenStreetMap Team (HOTOSM) survey data, and 3-arcsecond aggregated OSM building raster data sets. For the rural development pattern, the average building footprint area from 3-arcsecond aggregated OSM raster was used to determine the average total building area; assuming all rural development are single story. For all delineated development patterns, the HOTOSM in-situ building survey was used to establish the total building area by structure type per development pattern type using the surveyed building footprint and height values.

 

Replacement Cost:

Replacement cost was determined by profiling structural durability by development patterns, as determined remote sensing segmentation and digitization. Structural durability was classified into 3 tiers: temporary, semi-permanent and permanent. Examples: temporary housing in Tanzania include earthen, wood frame, and mixed materials; semi-permanent includes adobe block, unreinforced masonry; permanent includes reinforced concrete frame and reinforced masonry.

1) Construction Cost per building type (Residential): Replacement cost for permanent residential construction was available from one source or another for Tanzania, and although many structures did not fit this classification, their replacement cost, as indicated in the Resettlement Action Plans (RAPs) Report (2015), could be classified by durability, which generally correlated with the replacement cost provided. This intuitively makes sense, as people are likely to build the best house they can with the resources they have available- ultimately the quality or durability may be the best indicator of the price since detailed construction costs are not captured by the market economy.

2) Determining Building Type (Residential): The replacement costs for temporary and semi-permanent structures were gleaned primarily from the RAPs, and the replacement costs for permanent structures were taken from construction cost manuals. The replacement costs for each development pattern was determined by weighting by the mapping scheme for each structural type.

The project team developed a method based on a durability classification using the following categories:

a) Temporary – Structures classified as “temporary” are typically owner-built structures, characterized by mixed urban development and indigenous materials. No outside contracting (structural engineer, architect, developer, etc.) is required as construction is completed in the traditional and/or the most cost-effective way, with the assistance of friends, family and the community. Preparation for construction is minimal, as materials are locally sourced and processes such as forming adobe blocks or concrete blocks are typically not required. Utilities such as running water, electricity, etc. servicing individual structures is rare. Some structural descriptions in specific countries included rammed earth, earthen, mixed materials, light wood, mud walls and wattle & daub buildings.

b) Semi-permanent – Structures classified as “semi-permanent” are also owner-built structures, where construction is aided by friends, family and the community. As with “temporary” structures, outside professional contracting (engineering, architectural, developer, etc.) is rare or nonexistent, however construction is of higher quality and may require the assistance of skilled laborers (e.g. masons). Preparation prior to construction is generally required in the form of forming adobe blocks, concrete blocks, etc. Utilities such as running water, electricity, etc. servicing individual structures is rare. Examples included adobe block, unreinforced concrete block, unreinforced fired brick masonry, and confined masonry buildings.

c) Permanent – Structures classified as “permanent” are engineered structures (e.g. reinforced concrete frames, steel moment/braced frames, reinforced masonry bearing shear walls, etc.). The structure requires the input and knowledge from the likes of engineers, architects, developers, etc. and is typically built with skilled laborers. The construction quality is the highest of three types. Permanent structures are typically found in urban, industrial and “western” residential environments. Examples included reinforced concrete frame, reinforced masonry, reinforced concrete shear walls, non-ductile reinforced concrete frame with masonry infill walls, butler buildings and midrise steel structures.

3) Construction Cost Ratio (Residential): A ratio was calculated between the durability classes, expressed as a percentage of engineered construction. For example, semi-permanent construction for a given country may represent 40% of the total permanent construction cost, and temporary construction may represent 20% of the total permanent construction cost. The ratio approach is mathematically equivalent to taking the average value by durability class but has the added advantage of providing a basis for comparing the consistency of the approach across countries. Once these prices were established, a matrix of the expected durability class given development pattern was established for all structure types. Typically, these did not vary by development pattern, but in some cases, a structural class such as unreinforced masonry buildings encompass several types of quality and modifying the quality class by development pattern captured the range of quality and replacement cost. These matrices allowed mapping of the new replacement costs to structure class, and by cross-referencing the mapping schemes (percent structure type by country and development pattern), new estimates were produced for replacement cost by country and development pattern.

Below is the replacement cost by durability class:

Permanent materials: $339

Semi-Permanent materials: $141

Temporary materials: $61