The Sidi Rached bridge of Constantine is one of the most important landmarks and infrastructures of the city, it is used daily by the citizens as a mandatory passage; However, this can change soon if no actions are to be taken in the near future due to the critical conditions it is sat on; This bridge has been undergoing so many fixing works which were merely temporary solutions to the landslide occurring which is worsened by the soil erosion coming from natural factors( wind and water); In this paper, we address an Artificial Intelligence of Things (AIoT) based solution for remote monitoring and prediction of the coming hazards under the bridge foundation especially on the right bank. The method consists of using the a regression model for RUSLE and WEQ equation parameters calculation and future erosion estimation along with risk level classification with random forest model. For real life realization: The data that shall then be tested and classified is to be taken from the diferent sensors installed around the area.
Bridges are very crucial infrastructures when it comes to transportation [
To help ourselves react in the right way and time, it is necessary to have good view on the source of the problem which is considered to be the soil erosion under the right bank;This phenomenon is a global issue which technically is the efect of having soil particles transported due to natural causes like wind and water; [6] [7] where water soil erosion is more common and severe [8] [9]
It is worth to mention that diferent type of soil have diferent levels of erosion susceptibility because of their distinct geological formations because they afect in a direct and indirect way the soil erodibility factor K [6] [10] that is why every area and part needs to be studied on its own using the accurate data for better estimation and efective prediction.
Still; wind erosion needs to be taken in consideration especially seeing the fact that is estimated to increase by the coming years due to the climate change, when the wind intensity is stronger than the force maintaining the soil particles it drives them away causing erosion. For this matter as well; the soil texture plays a role in the level of wind erodibility, vegetation especially trees have high efectiveness in protecting the land against wind erodibility [11][12] [13]
As soil erosion represents a serious global issue so many researches and works addressed solutions to either predict, analyze or classify the coming hazards. When it comes to water erosion, a very common method is used which the Revised Universal Soil Loss Equation RUSLE; It consists of several factors related to the soil and its surroundings to estimate the amount of soil loss per year [14]
For instance; basing on the RUSLE, a work introduced new approaches like imagery, soil data and Artificial Intelligence to enhance the the soil erosion maps [ 15] Another work used ANN with several soil erodibility indices like Clay Ratio (CR )and modiefid clay ratio (MCR) to find the its probability. [ 16]
Also, in this study [17] the researchers used the GIS techniques and RUSLE equation to predict and estimate soil erosion risk. But to further dig into this formula; [18] focused on the calculating of cover factor basing on two subfactors, with phytoecological data along with the remote sensing GIS which helped mapping larger areas, this research found that the gravel-pebbles protection against erosion is even more efective than vegetation cover seeing the impact of the cover factor, [ 7] used an erosion mapping model to remotely detect the occurrence of erosion on a certain area by analyzing the vegetation cover through imagery; This method helps analyzing erosion phenomenon in a wide land in a short time. Others suggested the improvement of soil erosion prediction through analyzing the Land Use / Cover LULC change [9]
And although the RUSLE model seems to be linear, studies showed that it is hard to calculate the factors due to changing according to the regions and data [19] Hence why in this paper; we suggest the application of Artificial Intelligence because it finds its own parameters basing on constant data it keeps getting supposedly from the sensors installed around the area in question
AIoT as a technology has been used to solve multiple issues around the world in diferent aspects; When it comes to the Sidi Rached bridge risks, only temporary measures were given but no work has provided means to predict, estimate or classify any of the potential hazards facing it; Which is why hitherto we can say that this work is the first to exploit the benefits of AIoT to prevent upcoming safety menaces by predicting and classifying the erosion under the bridge foundation using the location specific features such as climate values and soil type; As well as being the first to use both water and wind erosion together to get the total loss and evaluate the coming risks in the following years according to the area’s climatic change pattern.
We can summarize this paper’s contributions as follows: • Application of IoT systems to collect data from the actual site for more specific analysis • Application of AI to analyze both the historical and actual data to predict any anomaly and upcoming hazard • Exploiting both water and wind erosion formulas for better estimation of the erosion level under the Sidi Rached bridge 4. Proposed Solution: seeing the common disregard of the wind erosion afect that can occur along with water erosion; we built in our solution in consideration of both phenomenons; Our solutions goes as follow: • Application of the linear regression model for future counting and estimating of the soil erosion due to the water efects • Application of the linear regression model for future counting and estimating of the soil erosion due to the wind efects These models would take in and train on data of the regular climate scenarios of the area around the right bank of the Sidi Rached bridge; learn the patterns and parameters and then estimate the annual soil erosion • The decision tree takes climate features and erodibility predicted values to be able to correlate the climate and weather factors with coming erosion for both wind and water and then estimate the risk level and alert before it happens • In realization of this work; diferent sensors mainly climate related (rainfall & wind) will give the data to the models which will help them estimate and learn the climate change patterns; this helps improving the accuracy and then give precise prediction of the coming hazards; the data does not need to be taken in real time in short periods and treated rapidly because the climate pattern has low progression speed.
We took in the values as follow according to the city of Constantine climate and estimation of area under the bridge conditions ( slope, vegetation...etc.) The data used is generated by programming given specific values and intervals K= Ranges from [0.02 to 0.1] due to the diferent component and level of clay/ sand and other geological properties in that area R= Ranges from [50 to 100] according to the city average precipitation LS=12 Estimated using the LS formula for slope and steepness of 30° and 70m C=0.5 P=0.25 Due to the small amount of natural vegetation and drainage system applied.
Seeing the fact that linear regression works with straight linear inputs and our tests included two variables (K and R) we opted for estimating the logarithmic value of A and then extract it reversely In the figures below; we display the ability of the model to quickly learn the pattern of the factors after training with corresponding data 4.1. Soil erosion due to water: using the RUSLE formula we get the equation: where • A = Estimated erosion or loss of material • R = Rainfall erosivity factor • K = Soil erodibility factor • LS = Topographic factor • C = Cover management factor • P = Erosion control practices = * * * * (1) 4.2. Soil erosion due to wind: where: • E = Soil loss due to wind erosion • I = Soil erodibility index • K = Surface roughness • C = Climate factor (wind speed and soil moisture) • L = Length of the field • V = Vegetation cover factor as performed with the RUSLE in the water erosion; we opted for the WEQ when it comes to wind erosion; = * * * * (2)
We fed the intervals of hazard classification to our model along with erosion prediction outputs the model to be be able classify the coming erosion hazards taking in consideration both features and their impact on the danger level. This phase can later help us predict the level of potential danger basing on the climate features which will impact the soil due to both wind and water.
• Training our model with the correspondent values as features • Passing the input tests to our pre-trained model to get its prediction of the classification • Testing the results of predictions with the real output test values with both accuracy_score and r2_score functions where both gave values higher than 0.98 corresponding to 98% accuracy 4.3.1. Random forest tree: This figure below represents the random forest tree.
This shows the process which the model goes by when learning and trying to make the most accurate prediction using the necessary features, i,e, the model takes final decision in favor on the highest vote in the tree branches This process shows to be an easy learn for this case scenario because of the input simplicity 4.3.2. The confusion matrix: This confusion matrix shows the ease the model has learning the patterns of the classification The confusion matrix of the random forest model shows how accurately it was predicting the classes where the: • rows represent the actual classes • columns represent the predicted classes • diagonal cells are the correct predictions made by the model • the rest of the cells are the mismatched; The wrong predictions 4.3.3. Final Classification: The down-below table shows the model’s attempt to classify the predicted data from both RUSLE and WEQ models after training. Those inputs were specifically input for diference classes to see the model’s ability to distinguish the features; with that the figure 9 shows the probabilities that model based on to make the decision
The previously shown results are mainly theoretical with predictions outputs as data due to our work limitations. Still it shows the ability and efectiveness of predictions and classifications
We repeated our work using the K_nearest neighbor model to test the diference of accuracy and decide which model would give us the best results After feeding the exact same data and evaluating the model with the same functions (r2_score and accuracy_score) we noticed that despite their high value, they were still lower than the accuracy reached by the random forest model In the down_below figure, we display the accuracy matrix of the KNN model after training and using it for the same data
Also, we tested the KNN trained model with the same new data to see the accuracy of its hazard level classification and we got the following output table along with the table of probabilities 5. Results observation and discussion: 1. Regression models can easily learn due to the simplicity of the inputs and patterns needed 2. When it comes to classification; random forest has better ability to learn and classify the potential hazard level of this context; This can be seen clearly in the both confusion matrices where the random forest matrix shows clear full ability to distinguish the classes when the KNN had less ease in that part 3. Good selection of models and their training data helps enormously in monitoring and predicting any coming danger before the bridge collapses and helps us react in the right or at least take measure to keep human life safe 4. Any wrong data or faulted estimation of any factor can lead to a completely wrong prediction ( false alarm or false negative) which may result in serious consequences 5. For better application of the proposed idea, real concrete data of the climate and weather features is needed to be fed to the random forest model to be able to correlate it with erosion values and then give better erosion prediction basing on weather condition and classify the potential hazard
In an important note; works like [20] and [18] have mentioned that adding gravel-pebbles and rock fragments increase the level of coverage even more than vegetation which helps decrease the soil erosion and stabilize the area as well as the implementation of AI can estimate climate scenarios that can occur in the future [21] which gives us the upper hand to see the potential threat long before they happen 7. Work limitations: 1. Despite the importance of the situation there is not much work and researches done around the area which leads to scarcity in the data needed especially soil properties related data 2. scarcity of the data decreases the models ability to give accurate prediction specific to that location 3. It also might be an expensive measure to get the right values for either coverage factor, soil properties and other indexes 4. Lack of data was an obstacle in a huge part to test the solution when it comes to features correlation and pattern extraction with erosion estimated value 5. Prediction of coming threats is not enough but react actions need to be taken and studied
The sidi Rached bridge is an important heritage for the city but in order to continue being in its position doing its major role it needs to be monitored constantly; The landslide and soil erosion in its right bank threatens its standing and people around it which is why we suggested the application of artificial intelligence of things to predict any coming hazards due to the soil erosion highly impacted by the climate change. Our work emphasizes the ease of prediction using the simple equations of soil erosion in the regression models; by taking these two factors it is easier to get a proper estimation of the soil erosion value; These values can be then learned in patterns and classified to diferent levels of hazard. Diferent classification models exist but the performance proved that this case study gives better results when applying Random forest model.
[3] M. Petrangeli, A. Viskovic, P. R. Marcantonio, et al., Rescuing the sidi rached bridge, in: IABSE Conference: Assessment, Upgrading and Refurbishment of Infrastructures, Rotterdam, The Netherlands, 6-8 May 2013, 2013, pp. 186–187. [4] M. A. Guemache, J.-L. Chatelain, D. Machane, S. Benahmed, L. Djadia, Failure of landslide stabilization measures: the sidi rached viaduct case (constantine, algeria), Journal of African Earth Sciences 59 (2011) 349–358. [5] C. S. R. BRIDGE, A. A. I. A. L. LANDSLIDE, Le pont sidi rached à constantine (algérie): Une culée dans un grand glissement de terrain, Bull. Serv. Géol. Nat. Vol 19 (2008). [6] A. Ekpo, L. Orakwe, E. Nwanna, D. Anizoba, C. Nwachukwu, Determination of soil erodibility (k) factor derived from diferent geologic formations of akwa ibom state, Nigerian Journal of Technology 40 (2021) 1096–1103. [7] A. Y. Saptari, A. Supriadi, K. Wikantika, S. Darmawan, Remote sensing analysis in rusle erosion estimation, Indonesian Journal of Geospatial 4 (2015) 34–45. [8] R. Lal, Soil erosion and the global carbon budget, Environment international 29 (2003) 437–450. [9] E. Mekonnen, A. Kebede, S. Asfaw, S. Feyissa, Optimizing soil erosion estimates of rusle model by analyzing land use/cover dynamics in upper awash river basin, central ethiopia, Geomatics, Natural Hazards and Risk 14 (2023) 2257363. [10] M. Liu, G. Han, X. Li, S. Zhang, W. Zhou, Q. Zhang, Efects of soil properties on k factor in the granite and limestone regions of china, International Journal of Environmental Research and Public Health 17 (2020) 801. [11] A. J. Parolari, D. Li, E. Bou-Zeid, G. G. Katul, S. Assouline, Climate, not conflict, explains extreme middle east dust storm, Environmental Research Letters 11 (2016) 114013. [12] H. Zhang, J. Fan, W. Cao, W. Harris, Y. Li, W. Chi, S. Wang, Response of wind erosion dynamics to climate change and human activity in inner mongolia, china during 1990 to 2015, Science of the Total Environment 639 (2018) 1038–1050. [13] M. T. Labiadh, J. L. Rajot, S. Sekrafi, M. Ltifi, B. Attoui, A. Tlili, M. Hlel, G. Bergametti, T. H.
Des Tureaux, C. Bouet, Impact of land cover on wind erosion in arid regions: a case study in southern tunisia, Land 12 (2023) 1648. [14] J. Van der Knijf, R. Jones, L. Montanarella, Soil erosion risk assessment in europe, 2000. [15] N. Samarinas, N. L. Tsakiridis, E. Kalopesa, G. C. Zalidis, Soil loss estimation by water erosion in agricultural areas introducing artificial intelligence geospatial layers into the rusle model, Land 13 (2024) 174. [16] S. Alqadhi, J. Mallick, S. Talukdar, M. Alkahtani, An artificial intelligence-based assessment of soil erosion probability indices and contributing factors in the abha-khamis watershed, saudi arabia, Frontiers in Ecology and Evolution 11 (2023) 1189184. [17] Y. Saadi, S. Mardiana, E. Pradjoko, Soil erosion prediction and risk assessment using rusle model and gis techniques in the nangka watershed, Journal of Water and Land Development (2022). [18] L. Boussadia-Omari, S. Ouillon, A. Hirche, M. Salamani, M. S. Guettouche, A. Ihaddaden, D. Nedjraoui, Contribution of phytoecological data to spatialize soil erosion: Application of the rusle model in the algerian atlas, International Soil and Water Conservation Research 9 (2021) 502–519. [19] R. Benavidez, B. Jackson, D. Maxwell, K. Norton, A review of the (revised) universal soil loss equation ((r) usle): With a view to increasing its global applicability and improving soil loss estimates, Hydrology and Earth System Sciences 22 (2018) 6059–6086. [20] J. Poesen, D. Torri, K. Bunte, Efects of rock fragments on soil erosion by water at diferent spatial scales: a review, Catena 23 (1994) 141–166. [21] H. Jain, R. Dhupper, A. Shrivastava, D. Kumar, M. Kumari, Ai-enabled strategies for climate change adaptation: protecting communities, infrastructure, and businesses from the impacts of climate change, Computational Urban Science 3 (2023) 25.