DRAI: a risk-based drought monitoring and alerting system in Brazil

Abstract

Drought is recognized as a devastating natural hazard, affecting human livelihood and causing a substantial economic impact. Consequently, experts and decision-makers concentrate on new approaches to reducing droughts’ economic and social effects through studies that focus on the monitoring, prediction, and risk analysis of drought to inform drought preparedness strategies and mitigation measures. This study presents the Drought Risk Assessment Interface (DRAI), a drought early warning system applied to the Brazilian semiarid region based on a composite index of meteorological drought risk. The risk index has two components: hazard and vulnerability. The hazard component considers meteorological indicators, while the vulnerability component encompasses social variables. Based on the opinion of experts from several countries, we define the weight of each of these indicators in the risk index using the analytical hierarchy process. Then, we propose a standard for generating warnings in the DRAI. The warnings are associated with seven drought risk mitigation measures validated by local technicians. We conclude that DRAI is a valuable tool to academics and practitioners, such as Civil Defences that can act directly in risk mitigation actions.

Appendices

Appendix A

This appendix discusses the validation of hazard and vulnerability indexes compared with other indices already developed in Brazil.

Hazard

The proposed methodology for the hazard index was validated by comparing the results of the hazard index in January 2019 with IDI and the Northeast Drought Monitor. The results were compared visually because the northeast drought monitor does not publish numerical data. Regarding the IDI, it was possible to compare the percentage of municipalities classified in the same category. See Fig. 

Fig. 6

Hazard index validation (January 2019). a Hazard index b IDI c Northeast drought Monitor

6.

According to the January IDI monthly report, moderate and severe drought conditions have increased since the previous month (December 2018), primarily in the Northeast region, with the states of Bahia and Pernambuco showing the most significant drought conditions (CEMADEN 2019).

In the IDI report, the municipalities with the most significant hazard intensity in January are in Paraíba, representing 0.2% of the municipalities, and presented a severe drought classification; they are Pilões, Pilõezinhos, and Pirpirituba. In the hazard index developed in this study, these municipalities have an increased degree of intensity. That is, they were classified as extreme drought. The second group of most significant hazards in IDI comprises 118 municipalities with moderate drought classification; 10% were classified as severe drought and 90% extreme drought. In this study, that is 1 or 2 degrees of difference in the hazard classification.

This difference in the degrees of intensity between the maps is because different variables are considered, different approaches are used, and the indexes are used in other places. The IDI is calculated for the entire national territory, while the index developed in this study and the Northeast Drought Monitor are used only in one region. Due to this difference in the application area, the reference of each index varies. The hazard level can be extreme in a semiarid municipality for the Northeast Drought Monitor and moderate for IDI since IDI also considers its methodology the Pantanal region, for example, suffers from intense droughts.

Thus, it is possible to observe that the IDI, the hazard index developed in this study, and the Northeast Drought Monitor show that the region between the north of Bahia and the state's border with Pernambuco has a higher level of drought. The Northeast Drought Monitor report supports this scenario. Precipitation in Bahia exceeded 100 mm west of the state, despite more than 300 mm forecasts. The results were less than 20 mm in the majority of the states. As a result, an improvement in the severity of the drought was insufficient, as revealed by the Vegetation Health Index (VHI), which showed that the drought was worsening.

The indices considered by the Northeast Drought Monitor are the SPI (for 3, 4, 6, 12, 18, and 24 months), which is based on the precipitation indicator, the SPEI, which is based on precipitation and potential evapotranspiration indicators (SPEI 2019) and the Standardized Runoff Index (SRI), which also considers the rainfall indicator (Martins et al. 2015). The indicators with the highest weights in the hazard index created in this study are precipitation and potential evapotranspiration. It explains why the severity level of this index and the hazard index developed in this study are similar.

The Northeast Drought Monitor was created by a group of experts and Brazilian institutions (federal and state entities, universities), with funding from the World Bank and international partnerships such as Mexico's National Water Commission (Conagua), the United States' Centre National Mitigation of Drought, and Spanish governmental and academic institutions. The National Water Agency (ANA) is the principal institution of the process in its most recent phase, with responsibility for federal coordination, articulation with the region's states, and growth to other states.

However, the final process of certifying the map involves civil society, which does not always have the appropriate information to assess the drought, despite living in the region daily. Furthermore, the map, which is its primary output, is hand-drawn. Because of these characteristics, the northeast drought monitor will be challenging to scale and automate. Nys et al. (2016) also pointed out these drawbacks.

On the other hand, IDI has no manual stages, making it scalable and automated and greater resolution that allows for a high level of monitoring for each municipality in the semiarid zone. The SPI, which is based on the precipitation indicator, the Vegetation Supply Water Index (VSWI) (Zhou et al. 2013a, b; Cunha et al. 2015; Alvalá et al. 2017), which combines the earth's surface temperature and the NDVI (Cunha et al. 2015), and the soil moisture are the indices used to create the IDI (CEMADEN 2019).

Experts validated in the questionnaire the use of physical data (precipitation, temperature, vegetation state, soil moisture, etc.) to produce indicators for measuring drought risk, stating that the most commonly used are SPI, SPEI, PDSI, Aridity Index (AI), VHI, and VSWI. It is feasible to find convergences of evidence of drought conditions using various indicators, which is necessary for enhanced decision-making consistency (Cunha et al. 2019a). Experts also proposed using geospatial technology, statistical modeling, sensitivity analysis, and run methods—a method based on statistical runs theory to analyze a time series sequentially (Nam et al. 2015)—to determine the duration and severity of droughts.

Another critical thing to consider is the data's source. The IDI and the Northeast Drought Monitor are based on data from various meteorological institutes, which might lead to inconsistencies between the two indices. To understand the differences between the databases, sensitivity analysis can be used to examine this kind of divergence.

Vulnerability

The municipalities with the highest vulnerability index in 2019 are Olho D’Água do Casado (AL), Palestina (AL), São José da Tapera (AL), Toritama (PE) and Poço Verde (SE).

The state of Pernambuco (PE) has 52% of the municipalities affected by drought and, according to the National System of Civil Defense and Protection (Letras Ambientais 2019), 53 municipalities in Pernambuco were recognized in an emergency in 2019 due to drought. However, only 24.8% of the municipalities have a contingency plan or hazard of natural disaster (Letras Ambientais 2019).

In Alagoas (AL), 34 of the 102 municipalities are affected by severe or moderate drought. According to the IBGE, from 2013 to 2016, Alagoas was the sixth state in the Northeast to register the most significant proportion of municipalities affected by drought (77.5%). Only 22.5% of cities have a contingency or drought prevention plan (Environmental Letters 2019).

Some initiatives in Brazil measure the level of vulnerability (IPEA 2015; Medeiros and Souza 2016), but few initiatives focus on vulnerability to drought. For the validation of the vulnerability index developed in this study, comparisons are made with two indices developed in Brazil, specific for droughts: the vulnerability index developed by INPE (INPE 2015) and the vulnerability index made from an initiative between the MMA, the Ministry of National Integration and WWF-Brazil (MMA 2017). See Fig. 

Fig. 7

Vulnerability index validation (2015)

7.

Regarding the index created by INPE, the indicators considered are different even though it considers the same vulnerability formula based on exposure, sensitivity, and adaptive capacity. The INPE map considers accumulated rainfall and average rainfall indicators for exposure indicators of the hazard component (UNISDR 2009). The sensitivity component considers data from the rural population, population employed in agriculture, poverty, rural households without access to water, rural households without adequate sanitation, reduction of agricultural production between 2011 and 2013, and reduction of livestock production between 2011 and 2012. According to the family's situation, the adaptive capacity component considers the diversification of the productive system, permanent private rural households with the existence of television and people aged ten and over, by literacy. In calculating the vulnerability, all these variables have the same weight.

It is also worth mentioning that the INPE map does not include the entire semiarid region, only a few municipalities in the northeast. Its objective was to map the vulnerability of the rural population in some municipalities in Paraíba, Ceará, Piauí, and Rio Grande do Norte. The municipalities considered the most vulnerable by the VI in 2015 are Olho D’Água do Casado (AL), Juazeirinho (PB) and Riacho de Santana (RN). Juazeirinho and Riacho de Santana are not included in the INPE map, but Olho D’Água do Casado has the same (very high) index as VI. Considering only the municipalities in common in the two studies, there are 333 cities. The average difference between the vulnerability index proposed in this work and INPE, both values between 0 and 1, is 15%.

The vulnerability index created by MMA (2017), called the Index of Vulnerability to Natural Disasters related to Drought (IVDNS), also uses the concept of exposure, sensitivity, and adaptive capacity. The following indicators are considered for exposure: increase in the average annual precipitation, increase in the standard deviation, increase in the power of droughts (from SPEI). It is worth mentioning that, like the INPE map, hazard variables, such as precipitation, are used to measure vulnerability (UNISDR 2009). For sensitivity, the following indicators are considered: land use, population percentage with an income lower than ¼ of the minimum wage, infant mortality, demographic density, and water demand and supply index. Finally, the indicators are considered for adaptive capacity: Municipal Human Development Index (HDI), social inequality, and illiteracy.

The IVDNS does not calculate the vulnerability exactly from a formula. It uses as a base period a 30-year time series of meteorological data (precipitation and temperature) for the years 1961–1990 (baseline), as instructed by the WMO. Then, projections are made from two climatic models: Eta-Hadgem ES 2 and ETA-MIROC 5, which represent the nesting of the regional Eta model, developed by INPE, with the respective global models.

Although the IVDNS maps have been transformed into tabulated values, for each municipality, validating the vulnerability index becomes a less assertive process due to: (i) IVDNS generates four indexes of vulnerability for each municipality (based on RCP 4.5 and RCP 8.5 for the models used Eta-Hadgem ES 2 and ETA-MIROC 5), which already makes the validation comprehensive and (ii), the IVDNS does not generate annual projections but for a period from 2011 to 2040.

In the report presented by MMA, the authors explain that the difference between the projections obtained by the models is a limitation of the study. Therefore, the results must be analyzed with caution. In addition, the authors mention that the large Brazilian territory makes it difficult to obtain results that are equally representative for all regions of the country. Therefore, for the index to be representative on more detailed scales, it is recommended that specific analyses are made for each area.

The MMA argues that the semiarid climate is more conducive to the low availability of water resources even though the effect of climate change is often pointed out as the primary determinant of vulnerability to drought disasters. However, a socioeconomic and management arrangement for water and soil intensifies the vulnerability of municipalities in the semiarid region.

The use of social data (such as political, economic, infrastructure, and social indicators) to create a composite index, the use of geographic information systems, the application of AHP, and surveys (done through interviews and data collection) were mentioned by the specialists who answered the questionnaire applied in this research as a method to assess vulnerability to drought.

Appendix B

This appendix shows the location of the municipalities mentioned throughout the paper Fig. 

Fig. 8

Municipalities

8.

Appendix C

Codes of the stations used: 82392, 82480, 82487, 82493, 82578, 82586, 82588, 82590, 82678, 82683, 82686, 82690, 82693, 82696, 82753, 82777, 82780, 82784, 82,791, 82792, 82795, 82797, 82798, 82879, 82882, 82886, 82890, 82893, 82975, 82,976, 82,979, 82983, 82989, 82990, 82992, 83076, 83090, 83097, 83179, 83182, 83,184, 83,186, 83,190, 83192, 83236, 83242, 83244, 83288, 83292, 83338, 83339, 83,344, 83,386, 83,393, 83,398, 83408, 83441, 83442 and 83483.

Codes of the faulty stations: 82287, 82474, 82583, 82594, 82691, 82789, 82870, 82986, 83088, 83221, 83295, 83388, 83389 and 83395.

Codes of the deactivated stations: 82294.