-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathmethodology_prioritization.html
241 lines (161 loc) · 9.48 KB
/
methodology_prioritization.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
<html>
<head>
<title>Prioritization of locations for off-grid rural electrification in Tanzania - Step by step guide</title>
<link rel="stylesheet" href="css/styles.css" />
<style>
body {margin: 2vw; /*vw=viewpoint width*/;
background-color:#F2EFE9}
//h1 {color: blue;}
//p {color: red;}
</style>
</head>
<body>
<div>
<h1>Prioritization algorithm</h1>
<h2>Tool development</h2>
<h3>1. Import of input data</h3>
<ul>
<li>Village polygons NBS</li>
<li>Grids from REA</li>
<li>CIESIN high resolution settlement layer (HRSL)</li>
<li>Global Settlement layer from JRC</li>
<li>Mini-Grid sites</li>
<li>Schools</li>
<li>Health sites</li>
<li>Mobile phone coverage</li>
</ul>
<h3>2. Streamlining data formats, extents and CRS</h3>
<h3>3. Analysis & Processing steps / Documentation on the spot for each step</h3>
<ul>
<li>Distance calculations from point locations to the grid</li>
<li>Grid length per village polygon</li>
<li>Overlaying the OSM village extracts with the village polygons</li>
<li>Decide on how and which population and village data to use (points vs. OSM polygons of residential areas vs. administrative village area polygons from NBS)</li>
<li>Additional village features, added to village information</li>
<ul>
<li>Schools</li>
<li>Health sites</li>
<li>Mobile phone coverage</li>
</ul>
<li>Exclusion criteria definitions and implementation</li>
<li>Ranking criteria definition: Implementation of filtering function accordingly</li>
</ul>
<h3>4. Output the prioritized locations</h3>
<ul>
<li>Webmap</li>
<li>Raw data formats (e.g. csv, json files) to upload to energydata.info</li>
</ul>
<h3>5. Upload metadata for each used data source</h3>
<br/>
<br/>
<h3>First insights into understanding the village data</h3>
<p>OSM + HRSL data show that many villages are well represented in OSM while some are not existing at all</p>
<p>Official village polygons indicate 17196 villages, OSM village data show x residential areas</p>
<p>The polygonised villages are sometimes covering one village in more than one polygon</p>
<img src="images/image004.png" width="605" height="292"> <br/>
<br/> <img src="images/image005.png" width="605" height="379">
<p>Resampling of the HRSL
to a lower resolution was done with the help of the warp function in QGIS, in
order to aggregate isolated pixels <br /> into more uniform populated surfaces. The
resampling was done with two algorithms (nearest neighbor (NN) and bilinear
interpolation (BI))<br />, each with 3 factors: 2, 5, 10 and 20. From the visual
inspection, the Bilinear Interpolation with factor 2 seems to give the best
results <br />(the ones that have the lesser data loss: view examples below). The
resulting file was then vectorized.</p>
<img src="images//image025.png" width="605" height="340">
<p>Figure 1 Resampled with NN, F=20</p>
<img src="images/image026.png" width="604" height="388">
<p>Figure 2 Resampled with BI, F=20</p>
<img src="images/image010.png" width="604" height="367">
<p>Figure 3 Resampled with NN, F=10</p>
<img src="images/image011.png" width="605" height="429">
<p>Figure 4 Resampled with BI, F=10</p>
<img src="images/image027.png" width="605" height="468">
<p>Figure 5 Resampled with BI, F=2</p>
<p>Figures 5 and 6 show that resampling by a factor of 2 shows good approximation of village extents, with minimum data loss, therefore this file was vectorized.(below)</p>
<img src="images/image014.png" width="375" height="301">
<p>Figure 6 Snapshot of vector file resulted from the BI with F=2</p> <br/>
<p>The vectorization still leads to 984942 settlement polygons which we are trying to simplify in
order to select and detect the village clusters with a reasonable size for
mini-grid development. Smaller structures, such as isolated farms, are more
likely to be electrified via stand-alone systems such as SHS.</p>
<p>For further insight into the population data, the raster file resampled with BI and F=2 was chosen and vectorized.</p>
<p>The next steps involve
processing of the different available datasets in order to better understand
them: buffering of the vector file and application of the zonal statistics
algorithm in QGIS, that shows statistical population data based on each polygon
as an output. The files used were: HRSL population layer (raster) and TZ
villages (vector file which includes census data).</p>
<p>After inspecting the
resulted vector file, the first conclusions are that there are various
differences between the HRSL data and the census data. For some villages the
difference could be up to -43000 or +23000 inhabitants (pictures below). Almost 5000 (out of 17196) villages have no census data.</p>
<p>Below, there are pictures of the extreme differences and one example of a village where the
difference between census and HRSL data (CFB) is 0.</p>
<p>For the village of Kanyenja, CFB is approximately -26000, most of it
concentrated in the northwest corner.Katindiuka is another example of a mismatch, where CFB is -44000, mostly concentrated in the
North.</p>
<p>The Wasa and Nkalama villages are the other extremes, where CFB is +23000 and +21000 respectively. Lastly,
Nayeme village, where CFB is 0. It can be seen that the polygon is completely covered by built-up area.</p>
<img src="images/image015.png" width="319" height="331">
<p>Figure 7 Village of Kanyenja</p>
<img src="images/image016.png" width="356" height="265"<br/>
<p>Figure 8 Katindiuka Village</p>
<img src="images/image017.png" width="326" height="321">
<p>Figure 9 Village of Wasa(Census-HRSL difference +23000)</p>
<img src="images/image018.png" width="321" height="300">
<p>Figure 10 Village of Mkalama (Census-HRSL difference is +21000)</p>
<img src="images/image019.png" width="604" height="343">
<p>Figure 11 Village of Nayeme</span> (difference is 0).</p>
<br/>
<h3>Improvement of grid data</h3>
<p>Clean grid data – the grid data contained data from 4 sources: AICD, WorldBank, REA and OpenStreetMaps.
However, WB and AICD data is not accurate, therefore was excluded from the analysis. Also, the 600 kV lines
in the dataset are most probably mislabeled.</p>
<p>Distance to grid was calculated from the centroids of each population cluster to the
grid. This was done with the help of different algorithms:</p>
<ol>
<li>V.distance for calculating the distance from each centroid to the grid</li>
<li>The resulting dataset was then joined with the original centroid dataset (based on the unique IDs)</li>
<li>Two new separate datasets were created, which specifically show which clusters are off grid and which are on grid. This was done with the help of the "select by location" algorithm, with the settings to select the clusters that do/not intersect with the grid, resulting in two shapefiles.
<ul>
<li>On grid shapefile shows 6778 clusters connected to the grid.</li>
<li>Off grid shapefile shows 28933 polygons that are not intersected by the grid, which shows a slight difference from the older analysis, with an outdaed grid dataset.</li>
</ul>
</li>
</ol>
<br/>
<img src="images/image020.png" width="557" height="403">
<p>Figure 12 Snapshot of cluster layer (on and off grid)</p>
<br/>
<p>The prioritization excel sheet contains population and surface area data based on
the unique id of every village. Population density data was also calculated.
Each ranking system (population, area, number of schools, health sites, water
access points) represents a number (0..1) which is basically the (population/pop density/area…) of the respective village relative
to the biggest number in the dataset. Considering weighting factors for each
criterion, a global ranking resulted, with “1” being Dar es Salaam.</p>
<br/>
<p>From the cluster shapefiles (centroids and polygons) a few entries were deleted, as they
were out of scope (some were outside the boarders of Tanzania; some were not
covering a village). However, the entries were saved in separate files. The two
polygon files (off and on grid) were merged again into one. The latest step is
to bring into the Cluster centroid file the information contained in the "TZ_village" vector file (village, ward, district and region
names, and global horizontal irradiance).</p>
<p>One other issue encountered in the data analysis was the fact that island cities and
villages were considered with no access to the grid (but they nevertheless have
access to electricity) and with a relatively big distance to the grid. This
caused problems in the rankings, so they were filtered out from the dataset.
Moreover, the sites with less than 100 inhabitants were also taken out of the
dataset, some of them being inaccurately classified as villages or out of scope
for mini grid development (see example below). They were saved separately, as a
shapefile and as a new tab in the prioritization excel file.</p>
<img src="images/image021.png" width="533" height="408">
<p>Figure 13 Example of a false positive</p>
<p>The Shapefile version has been cleaned of unnecessary information, such as number
of public buildings, health sites, etc. and except the grading system; also a
rank was implemented for ease of visualization. In the excel file, the second
tab is the legend for each column. This includes a short description, source
and date of download.</p>
</div>
</body>
</html>