Spatial analysis is the step that turns prepared geospatial data into answers a decision can rest on, combining vector and raster layers into a repeatable result rather than a one-off map. At Payeshgaran Farasanjesh Tabiat Fars this work follows data preparation and spatial database setup, and is designed around the agreed data model, positional accuracy and map scale. Outputs range from suitability surfaces and optimal routes to risk zoning, incident density maps and weighted layers feeding a spatial decision-support system. This page sets out the methods used, the inputs they require, the deliverables and the limits of each approach.
From Raw Data to a Well-Posed Question
Every spatial analysis rests on three foundations: a clear data model, one shared coordinate reference system, and geometric quality control. When layers are stored in different CRSs they must first be transformed to a common one; for local work in Iran UTM zone 39 (EPSG:32639) is common, while WGS84 serves data exchange. Point, line and polygon features also need clean topology so that later operations do not propagate hidden errors.
Typical inputs are vector layers in GeoPackage, Shapefile or GeoJSON; continuous rasters such as GeoTIFF and COG; and attribute tables joined to features through a shared key. Before any operation runs, the decision question is stated: what is measured, over which extent, against which criterion, and at what scale.
Delivery matters as much as computation. Results are handed over with metadata following ISO 19115, a written account of the processing steps and the model parameters, so that the analysis can be reviewed and updated rather than treated as a fixed artefact.
Overlay and Proximity Operations
Overlay is the core of vector analysis. Clip restricts data to the area of interest, Intersect builds the geometric and attribute intersection, Union retains all pieces, and Erase removes the unwanted part. Before these steps run, topology validation confirms that there are no unintended gaps, overlaps or dangling nodes.
Proximity operations convert distance into meaning. Buffer creates a geometric zone, while Near and Point Distance measure the actual distance to the closest feature. The distinction matters: Euclidean distance measures a straight line, whereas network distance measures a traversable path.
- Buffer and Multiple Ring Buffer for protection strips and impact zones
- Intersect and Union for layer combination and attribute transfer
- Clip and Erase for constraining the study extent
- Spatial Join for attaching attribute tables through a spatial relationship
- Near and Point Distance for adjacency and nearest-neighbour measurement
Network Analysis and Routing
Network analysis only behaves correctly on a line layer whose connectivity and travel rules are defined. For a street network, one-way restrictions, turn penalties, per-edge speeds and stop times must be entered into the model. Without these rules, the shortest path diverges from the operational route.
On that network, services such as shortest path, least-cost route, service areas and origin-destination matrices can be run. Vehicle routing problems arise when several vehicles must cover a set of stops under capacity or time constraints; an operating example is optimal routing of waste-collection vehicles for the Shiraz municipality waste management organisation.
Outputs are delivered as route layers, stop sequence tables and cumulative schedules, and can feed an operational dashboard rather than a static printout.
Site Selection and Suitability Modelling
Suitability modelling converts several heterogeneous criteria into one comparable index. The first step defines the criteria and their direction: some favour higher values, others lower. Each criterion is then mapped onto a common scale so that differing units do not distort the comparison.
Weighting can use simple weighted overlay or the analytic hierarchy process (AHP). In AHP a pairwise comparison matrix is built and its consistency ratio is checked; if the ratio exceeds the accepted limit the comparisons are revised. Constraint layers are applied separately as masks, for example steep terrain or flood-prone zones.
The model output is tested with sensitivity analysis to show how far the result depends on the chosen weights. Applied examples include siting livestock markets and nomadic settlement areas, and identifying suitable zones for a solar power plant.
- Define criteria and separate them from constraint layers
- Standardise criteria onto a shared scale
- Weight with weighted overlay or AHP, checking consistency
- Apply constraint masks to the final output
- Run sensitivity analysis on weights and criterion ranking
Interpolation, Terrain and Hydrology
When measurements exist only at scattered points but a continuous surface is required, interpolation comes into play. Common methods include inverse distance weighting, kriging and spline; kriging models the spatial structure of the data through a variogram and, where its assumptions hold, reports the estimation error as well. Method selection should be accompanied by cross-validation.
Terrain analysis underpins hydrological work. From an elevation raster, slope and aspect, hillshade, contour lines and a flow network are derived. Flow accumulation and watershed delineation resolve into individual stream channels, and indices such as topographic wetness support the ranking of zones.
This stage is bound tightly to the accuracy and spatial resolution of the elevation model; a poor raster can shift watershed boundaries and stream courses.
Hot-Spot, Density and Spatial Statistics
For incident data, kernel density shows intensity across a continuous surface while a hot-spot test such as Getis-Ord Gi* assesses the statistical significance of clusters. The two are complementary: one guides the eye, the other judges whether a pattern departs from randomness.
Global Moran’s I and local indicators of spatial association test spatial autocorrelation and show how strongly a value at one place depends on its neighbourhood. Interpreting these results requires stating the modifiable areal unit problem explicitly, since changing zone boundaries can change the outcome.
To model spatial relationships rather than describe them, locally weighted regression is available; unlike a global regression, it estimates coefficients separately for each location.
Uncertainty, Validation and Reproducibility
Every spatial result is as certain as its source data. Positional error in layers is measured as RMSE and reported against the map scale denominator; propagated through overlay operations, it can shift boundaries and influence analytical thresholds.
Reproducibility comes from documenting the model, recording parameters and retaining the data version. In decision-support systems the analysis chain should be re-runnable so that a new data release refreshes the result.
Limits are stated plainly: incomplete coverage over part of the extent, spatial resolution too coarse for the question, or missing valid attribute data can all weaken the confidence placed in the outcome.
Summary
Spatial analysis is a chain of data preparation, overlay and proximity, network analysis, suitability modelling, interpolation and statistical testing; each link depends on the previous one, and no clean result emerges without clean input.
At Payeshgaran Farasanjesh Tabiat Fars, this stage is delivered together with model documentation, final layers and metadata, so the result can operate inside a spatial decision-support system and be revisited in later cycles. Defining the question, the scale and the expected accuracy at the outset has the greatest influence on the quality of the final outcome.