Flood-susceptibility-based building risk under climate change, Hyderabad, India

AUROC

area under the receiver operating characteristic

BRT

boosted regression tree

CART

classification and regression tree

Cr

crores

DB

deep boost

DLNN

deep learning neural network

EAC

equivalent annual cost

FSBR

flood-susceptibility-based building risk

FSM

flood susceptibility modelling

GCMs

general circulation models

GEE

Google Earth Engine

GHMC

Greater Hyderabad Municipal Corporation

HEC-RAS 2D

Hydrologic Engineering Center River Analysis System 2D

HH

high building risk and high flood susceptibility

HM

high building risk and medium flood susceptibility

HRB

high-risk buildings

LH

low building risk and high flood susceptibility

LM

low building risk and medium flood susceptibility

LRB

low-risk buildings

MH

medium building risk and high flood susceptibility

ML

machine learning

MM

medium building risk and medium flood susceptibility

MODIS

MODerate Resolution Imaging Spectroradiometer

MRB

medium-risk buildings

NN

neural networks

NRSC

National Remote Sensing Centre

RCPs

representative concentration pathways

RF

Random Forest

SVM

support vector machine

TGPHMED

Telangana Public Health & Municipal Engineering Department

TSDPS

Telangana State Development Planning Society

XGBoost

eXtreme Gradient Boosting

Urban floods have been highly prominent natural disasters occurring in catchments across the globe. Some related factors are low-lying areas, clogged drains, settlements in flood-plain areas, and impervious surfaces. In addition, the widening gap between the increase in urbanization and available infrastructure is a major challenge. This inadequacy, in turn, will not be able to mitigate the impact of urban floods effectively. Another dimension is climate change, evidenced by high-intensity, short-duration, high-frequency rainfall affecting cities significantly (Nkwunonwo et al. 2020). This increases the quantity of water during most of the mentioned situations, which impacts the flooding area and escalates the flood susceptibility (Hammond et al. 2015).

Many floods have occurred in Hyderabad, Mumbai, Chennai, Bengaluru, and Vadodara. These cities have faced substantial financial losses and property damage (Flood Report 2021). In Hyderabad, severe floods occurred in 2000, 2006, 2016, 2018, and 2021. The flood depth range is 1–4 m in many locations (Rangari et al. 2021). Mumbai is another major city affected by urban floods due to climate change, as evidenced by the extreme rainfall of 944 mm for 24 h in 2005 (Sahany et al. 2010; Mumbai floods 2021). The case is similar with cities like Chennai, Bengaluru, and Vadodara, which are highly affected by urban floods (Vadodara floods 2019; Bengaluru floods 2021; Chennai floods 2021).

Numerous modelling approaches are employed to compute flood susceptibility. Machine learning (ML) is gaining momentum in classifying flood-susceptible regions due to its flexibility and adaptability (Baghbani et al. 2022; Saha et al. 2022). Shahabi et al. (2021) employed a deep belief–back propagation–genetic algorithm for Iran's Haraz watershed to generate flood susceptibility modelling (FSM). It was compared with other benchmarking ML techniques and found to be superior. Abedi et al. (2022) implemented the classification and regression tree (CART), XGBoost, random forest (RF), and boosted regression trees (BRT) to create an FSM of the Bâsca Chiojdului river basin. RF was found to be the best. Antzoulatos et al. (2022) employed RF, support vector machine (SVM), naive-based RF, and neural networks (NNs) to assess flood susceptibility in the Trieste, Monfalcone, and Muggia municipalities, northeast Italy. The RF model was highly rated with an F1 score of 0.99, followed by SVM, naive-based RF, and NN. Taromideh et al. (2022) reviewed ML applications to flood aspects at length. CART, RF, BRT, and several other models were employed to create an urban flood risk map for a case study in Iran. CART performed better than other ML algorithms.

General circulation models (GCMs) and associated representative concentration pathways (RCPs) are utilized for climate-based study. They help reproduce the historic observed climatic changes. Zennaro et al. (2021) studied the role of ML algorithms in climate change risk assessment and provided future directions. Chakrabortty et al. (2021) assessed flood susceptibility based on artificial neural networks (ANNs), deep boost (DB), and deep learning neural network in a climate change perspective for a case study in West Bengal, India. DB is the most preferred when compared with the others. They also studied flood susceptibility in detail in climate change scenarios. In summary, ML algorithms predict flood susceptibility but not flood depth (or building risk), which only hydraulic models can handle.

A few researchers have considered building risk an essential objective in vulnerability assessment for urban cities (Hossain & Meng 2020). It could be computed with inundation areas and flood depth information at the location. Park et al. (2021) studied flood risk assessment in Ulsan City, South Korea, using Hydrologic Engineering Center River Analysis System 2D (HEC-RAS 2D) for 2016. The inundated area was found to be 0.01–11.71 km², and the average flood depth for each administrative district was 0.47–1.20 m. A total of 20.6% of buildings were exposed to flood, resulting in high flood damage. Chen et al. (2022) assessed the flood risk map of Taiwan under RCP 8.5. It was concluded that approximately 14% of townships had high-risk buildings (HRBs). Ventimiglia et al. (2020) suggested waterproofing measures for the Mela River in northeastern Sicily, and Alabbad et al. (2022) for Iowa Middle Cedar Watershed to reduce property vulnerability and losses that would minimize building risk.

It is understood from the mentioned literature that both flood susceptibility and building risk are considered individually to judge the effect of flood. In this context, it is felt that by a combined index that considers flood susceptibility and building risk simultaneously, the impact of flood and possible mitigation measures can be understood holistically, which is the primary focus of this article. The authors propose flood-susceptibility-based building risk (FSBR) that addresses the mentioned challenge. Accordingly, the following objectives are formulated in the RCP 2.6 scenario:

• (i)

  Computation of FSBR

• (ii)

  Computation of cost of waterproofing that mitigates FSBR

The framework is applied to Greater Hyderabad Municipal Corporation (GHMC), India, due to its high population density and proneness to urban floods. The city has suffered major floods, and related details are presented in Table 1 (Rangari et al. 2021). Details of the case study are presented in the next section.

The GHMC area is 625 km² and is divided into various zones and circles (Greater Hyderabad Municipal Corporation 2023). The annual average rainfall is 840 mm and is at the maximum from June to September, leading to heavy flash floods. Average temperatures during winter and summer are 22 and 30 °C. The mean hottest and coldest months are May and December, respectively. The average annual rainy days and dry days are 56 and 265. The average maximum and minimum relative humidities in winter are 79% and 31%, whereas those in summer are 67% and 26%, respectively (TSDPS 2021).

The present article is a logical extension of the previous works of the authors (Madhuri et al. 2021a, 2021b), where information about modelling was discussed in detail for GHMC. Five ML algorithms, SVM, logistic regression (LR), K-nearest neighbour (KNN), AdaBoost, and XGBoost, are applied to understand the flood susceptibility of GHMC for historical data. Eight flood-influencing factors were used for this purpose. The area under the receiver operating characteristic (AUROC) is one of the standard methods for validating the model's performance (Tehrany et al. 2015). It is a graphical approach that describes the change in the algorithm's classification ability, as the probability threshold is altered (Shahabi et al. 2021). It compares and validates ML algorithms.

XGBoost is found to be the best, with a mean AUROC score of 0.83 and a standard deviation of 0.04. It is closely followed by AdaBoost, which has a mean AUROC score of 0.82 with a standard deviation of 0.04. Both algorithms significantly outperformed LR, SVM, and KNN with respective AUROC scores of 0.71, 0.74, and 0.77. Corresponding standard deviations are 0.07, 0.06, and 0.06.

Later, the study was analysed in climate change situations, as mentioned in the study area section. The flood susceptibilities generated from XGBoost for RCP 2.6 are classified into two different ranges, i.e., 30%–70% as moderate (M) and 70%–100% as high susceptibility (H), respectively. The 30% flood susceptibility value indicates a 30% chance of flooding in that location/pixel.

If building risk due to flood depth obtained by HEC-RAS 2D is high (H) and flood susceptibility obtained by ML algorithm is high (H), it is termed HH in the context of FSBR. It is high-priority combination that needs immediate action for mitigation measures by policy-makers. Suppose two buildings with high risk are situated in medium- and low-susceptibility areas; in that case, the priority is a medium-susceptible area. Six combinations of FSBR, namely, high building risk and high flood susceptibility (HH), high and medium (HM), medium and medium (MM), medium and high (MH), low and medium (LM), and low and high (LH) are employed to study the urban floods. Results related to flood susceptibility, building risk, and FSBR are discussed in the next section.

Table 3 presents zone-wise flood areas of buildings and roads under high, moderate, and low flood susceptibilities. Zones 12, 8, and 15 are highly susceptible, with exposed building areas of 14.61, 7.75, and 7.27 km², respectively. In the case of roads, zones 12, 5, and 13 are highly susceptible, with areas of 3.52, 3.11, and 2.65 km², respectively. In moderate susceptibility, buildings in zones 13, 15, and 12 are exposed in areas of 5.97, 4.74, and 4.29 km², respectively. In the case of roads, exposed zones are 12, 13, and 8, with areas of 6.41, 4.48, and 3.29 km², respectively. None of the zones have low flood susceptibility. Buildings and roads are highly susceptible in GHMC, with areas of 57.59 and 22.76 km².

Information about flood inundation mapping, flood depth, and building risk are presented in Table 4. The most vulnerable locations to flooding are situated near the Hussain Sagar and Musi River.

Zones 1 and 5 have high percentage inundation areas of 81% and 75% as they are near the Musi River and exist in low-lying areas. Zone 11 has the least percentage of inundation area of 61% due to its high elevation level. Zones 12 and 13 have flood depths of 0.3–8 and 0.2–6.8 m. This is due to rapid urbanization in these zones. Zones 1, 5, 8, and 15 are equally vulnerable, with flood depth ranges of 0.3–5.7, 0.2–5.8, 0.1–4.3, and 0.2–5.1 m. This may be due to their proximity to the Musi River and its low elevation of 466–525 m. The ranges of zone-wise percentages of HRB, MRB, and LRB are 38%–70%, 8%–18%, and 21%–49%, respectively. The highest number of HRBs was found in Zone 5, followed by Zone 1. It can be found that there is a greater number of LRB in zones 9, 3, and 14, as these are at relatively higher elevations. Inundated area, flood depth, and percentage of HRB, MRB, and LRB, respectively, GHMC-wise, are 442.53 km², 0.1–8 m, 51%, 13%, and 36%, respectively. The inundation area is 67%, more than half of the catchment area. This may be due to high-intensity, short-duration rainfall in the catchment.

In the HH category, zones 5, 15, 12, and 3 have affected areas of 13.07, 10.04, 8.50, and 7.64 km² with the corresponding number of buildings, 27,622, 30,850, 30,462, and 24,159, respectively. In contrast, the number of buildings in zone 9 is significantly less. This may be because of high elevation levels between 561 and 602 m. Zones 5 and 15 together occupied 36% of the total affected area. This may be due to their vicinity to the Musi River. In the HM category, zones 13, 12, 15, and 5 have 18,112, 16,148, 13,476, and 8,313 buildings with affected areas of 5.92, 4.85, 4.05, and 3.94 km², respectively. Zones 2, 7, 9, and 14 are the least affected.

In the case of MH, zones 12, 13, 15, and 8 have 11,102, 6,353, 5,996, and 6,411 buildings with affected areas of 2.25, 1.28, 1.22, and 1.09 km², respectively. Zones 12 and 13 have the highest curve number values, leading to high FSBR to flooding. In the MM category, zones 12, 13, 15, and 8 have 5,886, 4,732, 3,257, and 1,714 buildings with total affected areas of 1.36, 0.95, 0.66, and 0.46 km², respectively. Zones 2, 6, 9, and 16 have fewer buildings affected by floods as they are very far from the Musi River or at high elevation levels. Also, they have fewer pervious areas than zones 12, 13, 15, and 8.

In the LH category, zones 12, 15, 13, and 8 have 12,075, 6,652, 6,288, and 4,445 buildings with total affected areas of 2.28, 1.20, 1.18, and 1.11 km². Zones 12 and 15 occupy 25.1% and 13.83%, respectively, of the total affected buildings in this category.

Both these zones are more prone to flooding. In the LM category, zones 12, 13, 15, and 8 have total affected areas of 1.34, 0.88, 0.64, and 0.42 km² with 6,338, 4,794, 3,624, and 1,825 buildings. Almost 26.6% and 20.16% of the total affected buildings are in zones 12 and 13, respectively. In the HH, HM, MH, and MM categories, the maximum total affected areas in zones 15, 13, 12, and 12 are 10.04, 5.92, 2.25, and 1.36 km². The high slope of 19.13° for zone 15, areas near the Musi River, and the imperviousness of zone 12 are causes of urban flooding.

GHMC: In the case of the HH, HM, MH, MM, LH, and LM categories, the total affected areas are 63.40, 28.92, 9.52, 4.81, 9.26, and 4.56 km² (totalling 120.47 km²) with 182,178, 84,136, 46,238, 22,691, 48,092, and 23,781 buildings (totalling 407,116). The least number of buildings and flood-affected areas are in the LM category.

Only salient points are described here to understand the impact of waterproofing measures on the affected area and exposed buildings.

Waterproofing costs for zones 5, 15, and 12 for the HH category are 538.71 cr, 413.91 cr, and 350.38 cr; and costs for the corresponding EAC/building are Rs 27,800, 19,100, and 16,400 (refer to column 14, Table 5). Zones 2, 3, and 14 require less EAC/building of 13,900, 12,500, and 12,800 (refer to column 14). EAC/building for zones 13, 15, and 12 in the HM category are 19,200, 17,700, and 17,600 (refer to column 15). These values in the MH category are 11,900, 12,000, and 11,900 (refer to column 16); those in MM are 11,900, 11,900, and 13,600 (refer to column 17); those in LH are 11,000, 10,700, and 11,100 (refer to column 18); and those in LM are 10,800, 10,500, and 12,400 (refer to column 19). A decrease in EAC/building is observed in MH compared with HM due to there being less affected area.

Table 5

Number of buildings, waterproofing cost, and EAC/building for RCP 2.6

Zone no. (1) .	No. of buildings .						Cost (cr) Rs .						EAC/building Rs .						Total cost (Rs) (cr) (20) .	EAC/building/(⁠⁠) (21) .
	HH (2) .	HM (3) .	MH (4) .	MM (5) .	LH (6) .	LM (7) .	HH (8) .	HM (9) .	MH (10) .	MM (11) .	LH (12) .	LM (13) .	HH (14) .	HM (15) .	MH (16) .	MM (17) .	LH (18) .	LM (19) .
1	10,468	3,743	1,517	207	1,506	224	215.22	84.61	13.21	2.05	12.38	1.82	29,300	32,200	12,400	14,100	11,700	11,600	329.28	26,500
2	644	0	254	0	278	0	6.30	0.00	1.93	0.00	1.74	0.00	13,900	0	10,800	0	8,900	0	9.97	12,100
3	5,718	1,304	2,659	654	2,595	645	50.00	10.98	18.45	3.91	15.23	3.41	12,500	12,000	9,900	8,500	8,400	7,500	101.98	10,700
4	9,353	5,125	2,687	1,490	2,477	1,645	115.30	62.32	21.83	11.95	26.36	12.95	17,600	17,300	11,600	11,400	15,200	11,200	250.71	15,700
5	27,622	8,313	3,282	1,147	3,354	1,129	538.71	162.51	24.61	7.57	22.04	6.43	27,800	27,800	10,700	9,400	9,400	8,100	761.88	24,200
6	3,394	2,091	418	320	421	229	52.59	28.42	3.51	2.13	3.11	1.42	22,100	19,300	12,000	9,500	10,500	8,800	91.18	18,900
7	4,741	1,627	792	73	811	78	95.57	39.20	6.61	0.61	5.79	0.37	28,700	34,300	11,900	11,800	10,200	6,800	148.15	26,000
8	17,691	6,411	4,294	1,714	4,445	1,825	252.22	99.89	45.24	18.98	45.74	17.45	20,300	22,200	15,000	15,800	14,700	13,600	479.52	18,800
9	381	84	96	43	175	54	3.74	2.36	0.77	0.51	1.24	0.84	14,000	40,000	11,400	17,000	10,100	22,100	9.46	16,200
10	7,349	3,880	2,951	1,566	3,074	1,538	102.63	48.16	29.96	14.02	25.75	12.29	19,900	17,700	14,500	12,800	11,900	11,400	232.82	16,300
11	2,985	1,782	1,215	683	1,323	689	31.17	20.14	9.31	6.57	10.01	6.04	14,900	16,100	10,900	13,700	10,800	12,500	83.25	13,700
12	30,462	16,148	11,102	5,886	12,075	6,338	350.38	199.96	93.07	56.31	94.13	55.41	16,400	17,600	11,900	13,600	11,100	12,400	849.26	14,700
13	24,159	18,112	6,353	4,732	6,288	4,794	314.96	243.95	53.00	39.40	48.66	36.43	18,600	19,200	11,900	11,900	11,000	10,800	736.40	16,300
14	5,639	1,900	2,514	900	2,491	944	50.63	16.48	17.46	5.92	16.49	5.36	12,800	12,400	9,900	9,400	9,400	8,100	112.34	11,100
15	30,850	13,476	5,996	3,257	6,652	3,624	413.91	167.26	50.65	27.31	49.77	26.66	19,100	17,700	12,000	11,900	10,700	10,500	735.57	16,400
16	722	140	108	19	127	25	18.97	5.37	2.79	1.00	3.34	1.35	37,400	54,600	36,800	75,000	37,500	77,100	32.83	41,000
Total	182,178	84,136	46,238	22,691	48,092	23,781	2,612.31	1,191.60	392.40	198.24	381.80	188.25							4,964.6

Zone no. (1) .	No. of buildings .						Cost (cr) Rs .						EAC/building Rs .						Total cost (Rs) (cr) (20) .	EAC/building/(⁠⁠) (21) .
	HH (2) .	HM (3) .	MH (4) .	MM (5) .	LH (6) .	LM (7) .	HH (8) .	HM (9) .	MH (10) .	MM (11) .	LH (12) .	LM (13) .	HH (14) .	HM (15) .	MH (16) .	MM (17) .	LH (18) .	LM (19) .
1	10,468	3,743	1,517	207	1,506	224	215.22	84.61	13.21	2.05	12.38	1.82	29,300	32,200	12,400	14,100	11,700	11,600	329.28	26,500
2	644	0	254	0	278	0	6.30	0.00	1.93	0.00	1.74	0.00	13,900	0	10,800	0	8,900	0	9.97	12,100
3	5,718	1,304	2,659	654	2,595	645	50.00	10.98	18.45	3.91	15.23	3.41	12,500	12,000	9,900	8,500	8,400	7,500	101.98	10,700
4	9,353	5,125	2,687	1,490	2,477	1,645	115.30	62.32	21.83	11.95	26.36	12.95	17,600	17,300	11,600	11,400	15,200	11,200	250.71	15,700
5	27,622	8,313	3,282	1,147	3,354	1,129	538.71	162.51	24.61	7.57	22.04	6.43	27,800	27,800	10,700	9,400	9,400	8,100	761.88	24,200
6	3,394	2,091	418	320	421	229	52.59	28.42	3.51	2.13	3.11	1.42	22,100	19,300	12,000	9,500	10,500	8,800	91.18	18,900
7	4,741	1,627	792	73	811	78	95.57	39.20	6.61	0.61	5.79	0.37	28,700	34,300	11,900	11,800	10,200	6,800	148.15	26,000
8	17,691	6,411	4,294	1,714	4,445	1,825	252.22	99.89	45.24	18.98	45.74	17.45	20,300	22,200	15,000	15,800	14,700	13,600	479.52	18,800
9	381	84	96	43	175	54	3.74	2.36	0.77	0.51	1.24	0.84	14,000	40,000	11,400	17,000	10,100	22,100	9.46	16,200
10	7,349	3,880	2,951	1,566	3,074	1,538	102.63	48.16	29.96	14.02	25.75	12.29	19,900	17,700	14,500	12,800	11,900	11,400	232.82	16,300
11	2,985	1,782	1,215	683	1,323	689	31.17	20.14	9.31	6.57	10.01	6.04	14,900	16,100	10,900	13,700	10,800	12,500	83.25	13,700
12	30,462	16,148	11,102	5,886	12,075	6,338	350.38	199.96	93.07	56.31	94.13	55.41	16,400	17,600	11,900	13,600	11,100	12,400	849.26	14,700
13	24,159	18,112	6,353	4,732	6,288	4,794	314.96	243.95	53.00	39.40	48.66	36.43	18,600	19,200	11,900	11,900	11,000	10,800	736.40	16,300
14	5,639	1,900	2,514	900	2,491	944	50.63	16.48	17.46	5.92	16.49	5.36	12,800	12,400	9,900	9,400	9,400	8,100	112.34	11,100
15	30,850	13,476	5,996	3,257	6,652	3,624	413.91	167.26	50.65	27.31	49.77	26.66	19,100	17,700	12,000	11,900	10,700	10,500	735.57	16,400
16	722	140	108	19	127	25	18.97	5.37	2.79	1.00	3.34	1.35	37,400	54,600	36,800	75,000	37,500	77,100	32.83	41,000
Total	182,178	84,136	46,238	22,691	48,092	23,781	2,612.31	1,191.60	392.40	198.24	381.80	188.25							4,964.6

View Large

EAC/building for the HM category for zones 3 and 14 are Rs 12,000 and 12,400. These buildings have larger surface areas than those in the MH, MM, LH, and LM categories. Due to this, EAC/building is higher than in the mentioned categories (refer to column 15). Waterproofing costs (summation of all categories) of zones 5, 12, 13, and 15 are 761.88 cr, 849.26 cr, 736.40 cr, and 735.57 cr. The cost of the corresponding EAC/building is 24,200, 14,700, 16,300, and 16,400 (refer to columns 20–21).

GHMC: HH, HM, MH, MM, LH, and LM category buildings require an investment of 2,612.31 cr, 1,191.60 cr, 392.40 cr, 198.24 cr, 381.80 cr, and 188.25 cr (refer to column 8–13) and the total cost of waterproofing is 4,964.6 cr. On average, 17,400 as EAC/ building can be invested in preparedness for urban floods. As a note, HEC-RAS 2D provides limited information on high, medium, and low building risks. On the other hand, the FSBR has three options related to high risk (HH, HM, HL), medium risk (MH, MM, ML), and low risk (LH, LM, LL). This flexibility in decision-making makes FSBR preferable over individual usage of XGBoost and HEC-RAS 2D. A brief discussion to prove the potentiality of FSBR is as follows: HH covers 35% of the submergible building area and requires 2,612.31 cr for waterproofing. These values for HM are 16% and 1,191.60 cr, whereas HRB covers 51% and 3,808.91 cr. This gives an edge to policy-makers even in demarcating HRB into two categories and provides leverage while prioritizing the critical buildings for mitigation measures when the budget is limited. Similar inferences can be drawn for other categories.

There can be different ways for possible funding and its operations to encourage citizens to explore waterproofing measures, which are as follows:

• The tax benefits for implementing waterproofing measures can be provided to the house owner. This can be 1%–2% of the construction cost. This will encourage an individual to invest in waterproofing measures.

• Flood insurance/waterproofing bonds can be a better option for encouraging house owners to invest in waterproofing measures. This will enable people to safeguard their houses and families from a catastrophe.

• Use corporate social responsibility funds to safeguard public buildings against flood risk.

Urban floods have been highly prominent natural disasters occurring in catchments across the globe. This requires effective and sustainable mitigation mechanisms to minimize damage. In this context, a holistic approach, FSBR, is proposed by fusing XGBoost and HEC-RAS 2D (that provides building risk) in climate change situations and is applied to GHMC, India. The following conclusions are drawn:

• Buildings and roads have high susceptible areas of 57.59 and 22.76 km².

• The inundated area is 442.53 km², with ranges of flood depth being 0.1–8 m.

• Percentages of HRB, MRB, and LRB, respectively, are 38%–70%, 8%–18%, and 21%–49%.

• Total affected areas for HH, HM, MH, MM, LH, and LM are 63.40, 28.92, 9.52, 4.81, 9.26, and 4.56 km² totalling 120.47 km². Corresponding buildings are 182,178, 84,136, 46,238, 22,691, 48,092, and 23,781, totalling 407,116. The total cost of waterproofing is 4,964.60 cr.

This is the first-time application where ML and hydraulic modelling are complemented to derive an indicator, FSBR, for the urban catchment. The aim of this study is not to replicate the previous research works, but to give a new framework to researchers interested in this research area.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.