Figure 1. Conceptual framework A flowchart presents pathways from economic, social, and financial infrastructure to real G D P growth. The process begins with 3 infrastructure categories. Economic infrastructure leads to productivity, which then contributes to real G D P growth. Social infrast... More about this image found in Conceptual framework A flowchart presents pathways from economic, social...

Figure 2. Plot of CUSUM for coefficient stability A line graph presents CUSUM values from 10 to 20 against upper and lower 5 per cent significance limits. The CUSUM value starts near 1 at 10, decreases to about 0 at 12, rises slightly at 13, then falls to about minus 2.3 at 14. It subsequent... More about this image found in Plot of CUSUM for coefficient stability A line graph presents CUSUM va...

Figure 3. Plot of CUSUM square for coefficient stability A line graph presents CUSUM of squares values from 10 to 20 against 5 per cent significance limits. The CUSUM of squares value increases gradually from about 0.1 at 10 to about 0.2 at 13, then rises sharply to about 0.82 at 14 and abou... More about this image found in Plot of CUSUM square for coefficient stability A line graph presents C...

Figure 1. Example soil sensor A solar panel sits above the wireless gateway. The soil profile contains topsoil above subsoil. A soil moisture probe extends into the subsoil at 10 to 30 centimetres. More about this image found in Example soil sensor A solar panel sits above the wireless gateway. The s...

Figure 2. Study sites and climate zones A scatter plot maps 3 sites by longitude and latitude. The horizontal axis plots longitude in degrees east from 100.0 to 120.0. The vertical axis plots latitude in degrees north from 20.0 to 40.0. The temperate site, Chengdu, plots near 104.0 degrees eas... More about this image found in Study sites and climate zones A scatter plot maps 3 sites by longitude a...

Figure 3. Data processing and model pipeline A flowchart shows data inputs, preprocessing, feature engineering, model branches, concatenation, and water demand output. Meteorological data, soil moisture sensors, remote sensing imagery, and operational logs flow into data cleaning and preproces... More about this image found in Data processing and model pipeline A flowchart shows data inputs, prepro...

Figure 4. Descriptive statistical indicators of the sample: (a) meteorological data, (b) soil data, (c) vegetation growth indicators, and (d) operational data Four graphs compare meteorological data, soil moisture, vegetation indicators, and irrigation volume across 4 sites. Panel a plots prec... More about this image found in Descriptive statistical indicators of the sample: (a) meteorological data, ...

Figure 5. CNN-LSTM hybrid model architecture A flowchart shows a C N N and L S T M hybrid model architecture for water demand prediction. The left branch begins with input spatial data, remote sensing and images, followed by Conv 2 D with 32 filters, 3 times 3, ReLU, and max pooling, then Conv... More about this image found in CNN-LSTM hybrid model architecture A flowchart shows a C N N and L S T M...

Figure 6. Model validation results. Note: RMSE and MAE are reported in normalised units, and accuracy refers to prediction consistency within acceptable error bounds A combined bar chart and line graph compares R M S E, M A E, and accuracy across 3 climate zones. The horizontal axis lists ... More about this image found in Model validation results. Note: RMSE and MAE are reported in normalised...

Figure 7. Training results: (a) accuracy (%), (b) RMSE, (c) R², and (d) training time (s) Four bar charts compare accuracy, R M S E, R 2, and training time across rounds, models, and sites. Panel a plots accuracy in per cent against rounds. The values increase from 71.5 at round 1 to 82.3 at round 5, 89.2 at round 10, 92.4 at round 15, and 94.1 at round 20. Panel b plots R M S E for Our model, C N N, L S T M, and Random forecast with values of 0.147, 0.065, 0.032, and 0.019. Panel c plots R 2 for Our model, C N N, L S T M, L S T M, and Random forecast with values of 0.63, 0.74, 0.82, 0.83, and 0.89. Panel d plots training time in seconds for Site A, Site 15, Site C, Site C, and Site D with values of 92.1, 95.7, 92.4, 95.4, and 100. More about this image found in Training results: (a) accuracy (%), (b) RMSE, (c) R²...

Figure 8. Verification results Four bar charts compare accuracy, M S E, R 2, and calculation time across 4 sets. Panel a plots accuracy in per cent for Set 1 to Set 4. The values are about 89, 91, 92, and 91.5. Panel b plots M S E with values of about 0.042, 0.035, 0.027, and 0.027 for Set 1 t... More about this image found in Verification results Four bar charts compare accuracy, M S E, R 2, and c...

Figure 9. Evaluation results Four bar charts compare accuracy, M S E, R 2, and calculation time across 4 sets. Panel a plots accuracy in per cent for Set 1 to Set 4, with values of 89.5, 91.2, 92.1, and 91.7. Panel b plots M S E for Set 1 to Set 4, with values of 0.042, 0.035, 0.027, and 0.027... More about this image found in Evaluation results Four bar charts compare accuracy, M S E, R 2, and cal...

Figure 1. Study walls: (a) Wall A, (b) Wall B, and (c) Wall C; structures are shown highlighted in red Three panels show wall sections A, B, and C with marked survey areas. The panel a view shows Wall A with a long marked wall section containing graffiti. Panel b shows Wall B beside water, wit... More about this image found in Study walls: (a) Wall A, (b) Wall B, and (c) Wall C; structures are shown h...

Figure 2. Site map including TLS setup locations, viDoc scanning areas, and wall locations A site map shows wall locations, viDoc scanning areas, and T L S setup locations near Montlake cut. The map includes a metres scale from 0 to 60, a north arrow, Montlake cut, two T L S setup locations,... More about this image found in Site map including TLS setup locations, viDoc scanning areas, and wall lo...

Figure 3 viDoc setup: (a) Pix4D viDoc module, (b) lidar-enabled iPhone 14 Pro, and (c) DJI Osmos Gimbal A gimbal device holds an iPhone and viDoc attachment outdoors. More about this image found in viDoc setup: (a) Pix4D viDoc module, (b) lidar-enabled iPhone 14 Pro, and (...

Figure 4. Wall B data coverage: (a) P50 TLS cloud and (b) viDoc cloud, lapse in viDoc coverage highlighted in red, analysis window with sufficient viDoc data highlighted in yellow Two panels compare P 50 data coverage and viDoc data coverage. More about this image found in Wall B data coverage: (a) P50 TLS cloud and (b) viDoc cloud, lapse in viD...