Areal Interpolation in Python Using Tobler - Revision history

Smitch: /* Dasymetric Interpolation of Census Tracts and Dissemination Areas */

2022-01-07T15:22:35Z

Dasymetric Interpolation of Census Tracts and Dissemination Areas

Smitch: /* Introduction */ commented that the definition of interpolation is specific to a vector data model

2022-01-07T14:27:27Z

Introduction: commented that the definition of interpolation is specific to a vector data model

JohnFoster: /* Conclusions */

2021-12-24T02:20:38Z

Conclusions

JohnFoster: /* Step 4: Assess the Interpolation Results */ added images

2021-12-24T02:20:12Z

Step 4: Assess the Interpolation Results: added images

JohnFoster: /* Step 2: Count the Synthetic Population Points */ added images

2021-12-24T02:17:48Z

Step 2: Count the Synthetic Population Points: added images

JohnFoster: /* Step 4: Assess the Interpolation Results */ Added images

2021-12-24T00:44:39Z

Step 4: Assess the Interpolation Results: Added images

JohnFoster: /* Step 2: Inspect the Data */ added images

2021-12-24T00:39:54Z

Step 2: Inspect the Data: added images

JohnFoster: /* Data Sources */ fixed table format

2021-12-24T00:31:14Z

Data Sources: fixed table format

JohnFoster: Re-upload of page with main link fixed

2021-12-24T00:30:17Z

Re-upload of page with main link fixed

Show changes

@@ Line 366: / Line 366: @@
 ==== Dasymetric Interpolation of Census Tracts and Dissemination Areas ====
-Dasymetric interpolation, or masked area weighted interpolation, incorporates secondary information to help refine the spatial distribution of the variable. For example, the extent of census tracts may include large portions of land that are uninhabited (e.g. parks, fields, water, etc.). Secondary information about the census tracts' land cover is used to mask out those uninhabited areas which don't contain the extensive variable. Area weighted interpolation is then applied to the masked census tracts and the results should be more accurate than simple area weighted interpolation on its own. This can be done by clipping the census tracts' using another set of appropriate polygons(e.g. land cover polygons, municipal zoning, etc.) and then running the areal interpolation function, but Tobler includes a function to do this automatically using a land cover raster as the secondary information. Please refer to Tobler's documentation of the <code>tobler.dasymetric.masked_area_interpolate()</code>](https://pysal.org/tobler/generated/tobler.dasymetric.masked_area_interpolate.html#tobler.dasymetric.masked_area_interpolate) function for help with the parameters.
+Dasymetric interpolation, or masked area weighted interpolation, incorporates secondary information to help refine the spatial distribution of the variable. For example, the extent of census tracts may include large portions of land that are uninhabited (e.g. parks, fields, water, etc.). Secondary information about the census tracts' land cover is used to mask out those uninhabited areas which don't contain the extensive variable. Area weighted interpolation is then applied to the masked census tracts and the results should be more accurate than simple area weighted interpolation on its own. This can be done by clipping the census tracts using another set of appropriate polygons(e.g. land cover polygons, municipal zoning, etc.) and then running the areal interpolation function, but Tobler includes a function to do this automatically using a land cover raster as the secondary information. Please refer to Tobler's documentation of the <code>tobler.dasymetric.masked_area_interpolate()</code>](https://pysal.org/tobler/generated/tobler.dasymetric.masked_area_interpolate.html#tobler.dasymetric.masked_area_interpolate) function for help with the parameters.
 The land cover raster that we will use comes from the [https://open.canada.ca/data/en/dataset/4e615eae-b90c-420b-adee-2ca35896caf6 2015 Land Cover of Canada]. This 30 m spatial resolution land cover product is produced every 5 years and covers the whole country. Tobler's dasymetric interpolation function will automatically clip the raster to the extent of the source data (City of Ottawa) but I have already clipped it to reduce the file-size and processing time.
@@ Line 423: / Line 423: @@
 <span id="step-4-assess-the-interpolation-results"></span>
 === Step 4: Assess the Interpolation Results ===

@@ Line 4: / Line 4: @@
 = Introduction =
-Areal interpolation is the process where the attributes of one set of polygons are interpolated to a different set of polygons. There are many applications for this process, for example, the interpolation of census data from census tracts to neighborhoods, or the interpolation of aggregated environment data from one set of geographic boundaries to another.
+Areal interpolation, in vector-based GIS data models, is the process where the attributes of one set of polygons are interpolated to a different set of polygons. There are many applications for this process, for example, the interpolation of census data from census tracts to neighborhoods, or the interpolation of aggregated environment data from one set of geographic boundaries to another.
 Tobler, a component of the [https://pysal.org/ Python Spatial Analysis Library] (PySAL) is a package of Python functions designed to perform and support areal interpolation. Visit Tobler's [https://pysal.org/tobler/index.html website] and [https://github.com/pysal/tobler github page] for the official documentation.

@@ Line 1,318: / Line 1,318: @@
 <span id="conclusions"></span>
-== Conclusions ==
+== Conclusion ==
 Looking at the statistical and visual representations of the synthetic results we can see a familiar pattern emerge. The error and correlation values show that when choosing between census tracts and dissemination areas the latter is the better choice. Also, when comparing between the two different approaches to interpolation, the dasymetric method is superior that the simple area weighted method. Bear in mind, the model in Part 2 is quite over-fit as we used the same land cover raster for the dasymetric interpolation and to create the bounds within which the synthetic points were generated - in reality there would quite a bit more variation. Despite this issue, I believe we can still conclude that using the finest spatial resolution of source data, whether through the choice geometries, or through the contribution of secondary information such as land cover, will give the best results when performing areal interpolation.

@@ Line 1,148: / Line 1,148: @@
 </div>
+[[File:11_sample_merged_results.png]]
 <div class="cell code">
@@ Line 1,199: / Line 1,201: @@
 </div>
+[[File:12_stats.png]]
 <div class="cell code">
@@ Line 1,244: / Line 1,248: @@
 </div>
+[[File:13_facet_map.png]]
 <div class="cell code">
@@ Line 1,306: / Line 1,312: @@
 </div>
+[[File:14_1to1_plot.png]]
 <div class="cell markdown">
 <span id="conclusions"></span>
 == Conclusions ==

@@ Line 976: / Line 976: @@
 </div>
+[[File:8_ct_points_count.png]]
 <div class="cell code">
@@ Line 996: / Line 998: @@
 </div>
+[[File:9_da_points_count.png]]
 <div class="cell code">
@@ Line 1,016: / Line 1,020: @@
 </div>
+[[File:10_nbhd_points_count.png]]
 <div class="cell markdown">
 <span id="step-3-areal-interpolation-of-the-synthetic-population-data"></span>
 === Step 3: Areal Interpolation of the Synthetic Population Data ===

@@ Line 461: / Line 461: @@
 </div>
+[[File:3_sample_merged_results.png]]
 <div class="cell markdown">
@@ Line 566: / Line 568: @@
 </div>
+[[File:4_stats.png]]
 <div class="cell markdown">
@@ Line 636: / Line 640: @@
 </div>
+[[File:5_facet_map.png]]
 <div class="cell markdown">
-That's a pretty slick set of maps, if I can say so myself! Each map can be explored by changing the zoom level and by hovering the cursor over different features.
+That's a pretty slick set of maps, if I can say so myself! Each map can be explored by changing the zoom level and by hovering the cursor over different features (if you're viewing the interactive Jupyter notebook version of this tutorial).
 There are some other limitations with Plotly though. While it does support tiled base maps through their MapBox figures (e.g., <code>px.choropleth_mapbox()</code>) it's not possible to create facet plots (i.e., side-by-side plots) with a tiled base map. Also, Plotly doesn't make it easy to represent the values as discrete classified bins instead of a continuous color scale. Instead, in the map above I've simply limited the <code>Error (%)</code> color bar range to 100 despite some neighborhoods having higher percent errors higher than 100. Feel free to adjust the <code>rangecolor=</code> argument to a different scale, if you so desire.
@@ Line 666: / Line 672: @@
 </div>
+[[File:6_explore_map.png|1200px]]
 <div class="cell markdown">
@@ Line 739: / Line 747: @@
 </div>
+[[File:7_1to1_plot.png]]
 <div class="cell markdown">
@@ Line 757: / Line 767: @@
 <span id="part-2---areal-interpolation-of-synthetic-population-data"></span>
 == Part 2 - Areal Interpolation of Synthetic Population Data ==

@@ Line 224: / Line 224: @@
 </div>
 <div class="cell markdown">
+[[File:1_compare_random_samples.png]]
 From this we can see that the census tracts and neighborhoods have roughly similar population values but the dissemination areas are quite a bit smaller.
@@ Line 256: / Line 257: @@
 </div>
+[[File:2_plot_the_geometries.png|1200px]]
 <div class="cell markdown">
@@ Line 300: / Line 303: @@
 <span id="step-3-areal-interpolation-of-census-data"></span>
 === Step 3: Areal Interpolation of Census Data ===

@@ Line 87: / Line 87: @@
 The data for this tutorial has been preprocessed for your convenience and can be found in <code>data/</code>. All of this data comes from free and open sources. Links to the original geometry and attributes can be found below along with notes regarding the preprocessing that was performed.
+{| class="wikitable"
-{|
 !width="20%"|
 !width="20%"| Geometry source
@@ Line 136: / Line 136: @@
 <span id="areal-interpolation-in-python-using-tobler---tutorial"></span>
 = Areal Interpolation in Python Using Tobler - Tutorial =