Introduction

The Aznalcóllar mine complex is located in southwest Spain (Fig. 1a), in the western foothills of the Iberian pyrite belt (Borja et al. 2001), which has been mined since the Roman Empire (CSIC 2008). On 25 April 1998, the pond where the mine tailings had been stored since 1978 collapsed. Millions of cubic meters of a complex fluid were spilled directly into the Agrio River and subsequently into the Guadiamar River, affecting the area’s environment. The spill reached the northern limit of the Doñana National Park (CMA 1998), where the flow was stemmed by the efforts of the local and national administrations (AGE and JA 1999).

Fig. 1
figure 1

Pond location and characteristics. a Geometrical particularities and general location of the pond (red dot). b Sketch of the pond (plain view) and cross-section type of the east dyke of the pond (source: adapted from Alonso and Gens (2006a)). c Evolution of representative cross-section (source: adapted from Alonso et al. (2010), Gens and Alonso (2006), and Zabala and Alonso (2011)). *The external slope varied from 1.75H:1V at first construction stages to 1.2H:1V just before the failure

Full size image

The event was one of the most catastrophic ecological mine disasters of the Iberian Peninsula and Europe (CSIC 2008). At the time, it was the second largest in the world by volume (Turner et al. 2002), and the largest reported in Europe (Nikolic et al. 2011), and it remains the fifth largest spill worldwide (WISE 2020). When the event happened, media reported extensively on the disaster. It was first reported by the regional and national administrations, and then by national and international daily newspapers, radio and television newscasts, and special programmes (Elías 2001). Scientific publications started to appear a short time after the catastrophe, with the aim of clarifying the causes and consequences of the event.

Currently, there are more than 400 research articles on the disaster (Madejón et al. 2018b). The event has also motivated two special issues (Grimalt and MacPherson 1999; IGME 2001) and two reviews (Ayala-Carcedo 2004; Madejón et al. 2018b). Although well-documented, with published studies by many scientific disciplines, some fundamental aspects of the spill itself have not yet been analysed in detail (Ayala-Carcedo 2004). Only 11% of the publications on it are classified as “Water research” (Madejón et al. 2018b), and no reviews have been found in the literature describing in detail the hydraulics of the flow that propagated from the pond failure. Although there seems to be a general agreement about some of the essential hydraulic aspects, such as for example the spilled volume (5–6 hm3) (e.g. Arenas et al. 2001; Madejón et al. 2018b; Pain et al. 1998; Rodríguez-Tovar and Martín-Peinado 2009) and the fluid type and behaviour (mud-like) (e.g. Benito et al. 1999; Guerrero et al. 2008; Hernández et al. 1999; Madejón et al. 2018a), significant uncertainties regarding the hydraulics of the spill persist: the volume released, the length and area affected by it, the nature of the fluid spilled, how the fluid left the pond, and its propagation velocity, among others.

The literature contains a wide range of values regarding the variables that characterized the phenomenon, though few studies have attempted to reproduce the flood using numerical methods (see “Discussion”). Only a few researchers have questioned if the pond volume was enough to affect the area it did (Ayala-Carcedo 2004; Padilla et al. 2016; Turner et al. 2002), and none have looked for other water sources (e.g. overland flow) that entered into the area after the event and, in such case, how they might have interacted with the spill and distorted the volume ultimately recovered.

The main goal of this paper is to review the most relevant hydraulic aspects of the Aznalcóllar mine disaster. It addresses the pond and fluid characteristics, the break type, the breaking time, the flow propagation, the volume potentially stored and subsequently spilled, area dimensions affected by the spill, and the morphological changes carried out during the clean-up activities and due to the environmental re-naturalization plans promoted by the national and local authorities. New data are presented and analysed to clarify the most controversial aspects of the hydraulics of the mine disaster. For these purposes, we compiled information about the post-failure topography, including the pond area, analysed the gauge station data and flood velocity, assessed the fluid behaviour using video observation-based classification techniques, and examined the most relevant aspects that could condition the numerical reproduction of the event.

The document has been structured as follows: first, a “Reported Facts” section for contextualization; second, a section on the “Reported uncertainties” found in the literature; and, third, “New evidence” and data that appear to clarify some of the controvertible aspects. The “Discussion” and “Conclusions” highlight some of the most relevant deductions from the literature review together with our analysis of the new evidence, some of which point out some discrepancies in the previously published work.

Reported Facts

Construction and Filling Process

Mine waste was deposited in a pond to the south of the open-mine pit called ‘Los Frailes’, which was very close to the Agrio River, and which was composed of two lagoons (Fig. 1a). The final structure was planned to be built in 20 stages, starting in 1978 and continuously increasing in height and storage capacity. It was at stage 16 when the disaster occurred (CMA 1998).

A schematic construction process was first described by Gens and Alonso (2006). These authors focused on geotechnical aspects (mainly the dyke height), not mentioning some aspects of the initial geometry of the pond and the increased storage capacity during the construction process. According to this publication, the pond was designed with a cross-section (Fig. 1b) defined by a 1.9H:1V internal slope and a 1.75H:1V external slope (Alonso et al. 2010; Zabala and Alonso 2011). However, two changes modified this geometry: (1) in 1985, the slope of the external dyke of the east side was changed to 1.24H:1 V (Alonso et al. 2010; Gens and Alonso 2006), although according to Ayala-Carcedo (2004) this slope was slightly higher, 1.2H:1V (Fig. 1c); and (2) in the period 1988–1990, the crest width was increased from 14 m (design value) to 36.5 m.

The area occupied by the fluid increased with the filling process, which depended on the mine management. No published data are available regarding the filling process, though.

The pond was originally projected to store 32.5 hm3 (stage 20 of the original project), but when the dyke broke, it was at stage 16. Thus, at failure, the stored volume must have been less than the projected one.

Fluid Type

The properties of a fluid are essential to understand its dynamic and static behaviour. However, a detailed description of the fluid deposited in the pond is presently not available. Although the fluid can generally be classified as tailings, the residuals of the Aznalcóllar mine activities were mainly composed of fine solids and water with dissolved metals (CMA 1998). Mine tailings were separated in two lagoons: the northern with pyroclastic tailings and the southern with pyritic tailings (Fig. 1a). The two are commonly termed in the literature as “acid waters” and “muds,” respectively, and the same terminology will be used herein.

At this point, two types of fluids can be inferred: a water-like fluid for the “acid waters” (pyroclastic tailings) and a mud-like fluid for the “muds” (pyritic tailings); thus, two different fluid behaviours are expected. However, when the failure occurred, both fluids were spilled and mixed; thus, their properties probably changed (Ayala-Carcedo 2004). Several authors did not split the fluid characteristics and considered them as a unique fluid.

Breaking Type and Time

The type and causes of the embankment rupture are well described from a geotechnical point of view (Alonso and Gens 2006a, b; Gens and Alonso 2006). The pond failure was initiated at the east dyke of the southern lagoon by a 600 m long fan-like displacement that achieved a maximum movement of 55.2 m. According to these authors, the failure might have occurred in less than 16 s, immediately affecting the east dyke of the northern lagoon and the central dyke, and generating an opening in the embankment.

The incident happened in the early hours of Saturday, 25 April, but at an unknown time. At 3:20 a.m., an neighbour reported to the authorities of Sanlúcar la Mayor that a strong din was heard, the Guadiamar River had overflown, and the water had an unusual colour (AGE and JA 1999). The place where the observations took place was not disclosed, but it is clear that the flood had already started.

Spilled Volume

Volumetric data is very controversial, from the storage capacity of the pond before the pond failure to the volume registered at the El Guijo gauge station. The exact geometry of the pond, the filling process, and the volume stored just before the disaster are still unknown. The limnograph recordings (Fig. 2b, dash line) registered at a gauge station (Fig. 2a, location #1) and an approximation of the recovered volume are partially known.

Fig. 2
figure 2

a Location of the pond failure (red dot) and the particular observation places (denoted with a number on the left side of the Guadiamar River). b Limnograph registered at EA90 (location #1 of (a)) on 25 and 26 April, according to Borja et al. (2001) (continuous line), to Consultec Ingenieros (1999) (dot line) and to the real measurements (dash line)

Full size image

Spill Wave Propagation

There is a general lack of data regarding the propagation of the spill, besides discrepancies in the distances from the breaking point to the places where the flood front was observed. The flood was only registered at the El Guijo gauge station (EA90), 7.1 km downstream of the breaking point (Fig. 2a, location #1). By contrast, Borja et al. (2001) and Ayala-Carcedo (2004) indicated that the gauge was 11 km downstream of the pond, and this could have affected estimations of the flow propagation velocity (as denoted later).

Affected Length

The spill traced a path from the Agrio River to the Guadiamar–Guadalquivir junction (Fig. 3a), flooding the riverbeds and the floodplains of both rivers. It flooded a 5 km long reach of the Agrio River, 1.9 km of it being upstream of the breaking point. Due to the backwater effect (Castelltort et al. 2020), 2.2 km of the Guadiamar river were also flooded upstream of the Agrio–Guadiamar junction (Fig. 3b). By means of the same effect, 3.9 km of the Brazo de la Torre were affected by the spill downstream of the Entremuros zone (Fig. 3d). The spill finally reached the Guadalquivir River (Fig. 3d), flooding more than 80 km of the Agrio, Guadiamar, and Guadalquivir rivers (Fig. 3a).

Fig. 3
figure 3

a Affection of the spill. a Extent of the spill including muds (brown) and acid water (green), and the potential mixture zone (brown-green dashed area). Aerial image of the affected area taken five days after the spill: b from the pond to the Agrio-Guadiamar junction; c from the Don Simón ford to the Los Vaqueros ford; d junction of the Brazo de la Torre and Guadiamar River (red polygon highlights the limits of the affected area) (source: 1:10,000-scale flight from JA (2003))

Full size image

Affected Area and Deposited Muds

The extent of the spill affected not only the riverbed but also the riverbanks. The spill was partially retained in the Entremuros area (Fig. 3a), where the valley widens, and the velocity of the fluid slowed down, causing the suspended particles to be deposited.

Recovered Volume

The clean-up activities were split in three compartments: from the pond to the Sanlúcar la Mayor bridge (Fig. 2a, location #2); from this point to the Don Simón ford (Fig. 2a, location #6); and from this point to the end of Entremuros (Fig. 2a, location #9). The cleaning of each area was conducted by the mining company, the Confederación Hidrográfica del Guadalquivir (CHG), and the local administration, respectively (AGE and JA 1999; Antón-Pacheco et al. 2001; CSIC 2008).

However, recovery of the muds was not limited to the original topography; the natural terrain was also removed to ensure a full clean-up. According to AGE and JA (1999), the extracted natural soil thickness could have varied from a few centimetres up to 0.1 m. However, in the area allotted to the mining company, the terrain was dug deeper, causing several morphological changes (Antón-Pacheco et al. 2001).

Reported Uncertainties

Construction and Filling Process

At the end of the construction process, the pond height would have reached 25 m (CMA 1998), being 21 m just before the failure (stage 16). However, other values exist in the literature, ranging from 16.4 (Castro-Díaz et al. 2008) to 30 m (WWF/Adena 2008). The most referenced values are 27 and 28 m (Alonso and Gens 2006b; Alonso et al. 2010; Ayala-Carcedo 2004; Gens and Alonso 2006; Gómez de las Heras et al. 2001; McDermott and Sibley 2000; Penman et al. 2001; Zabala and Alonso 2011). According to Gens and Alonso (2006), the crest elevation in 1998 was 68.5 m (Fig. 1c), 24.8 m above the foundation, the bottom of the pond being at ≈ 43.7 m. No explicit references about the foundation depth are found in the literature.

The first referenced value of the area occupied by the fluid before the break is 140.8 ha (Simón et al. 1999). This is slightly smaller than the one presented by Consejo Superior de Investigaciones Científicas, CSIC (2008), and Santofimia et al. (2011, 2013), of 150 and 170 ha, respectively. Higher values (200 ha) are reported as well (Arenas et al. 2001; Ayala-Carcedo 2004; Gómez de las Heras et al. 2001; McDermott and Sibley 2000; WWF/Adena 2008). Ayala-Carcedo (2004) split the total area into the northern and southern lagoons, being 150 and 50 ha, respectively.

According to CMA (1998), the pond volume could have been 26.5 hm3 just before the break. Other authors also support similar values of 26 hm3 (Arenas et al. 2001, 2008; Borja et al. 2001; CSIC 2008) and 25 hm3 (Manzano et al. 2000; Santofimia et al. 2011). In contrast, McDermott and Sibley (2000) indicate that the pond stored 15 hm3 of fluid at the time of the failure, while Ayala-Carcedo (2004) suggests that the volume ranged from 15 to 20 hm3. Other authors (AGE and JA 1999; Cabrera 2000; Grimalt and MacPherson 1999) used the final projected volume of 32.5 hm3 as the stored fluid volume just before the disaster.

Fluid Type

Properties of the fluid were first mentioned in the official report of Administración General del Estado and Junta de Andalucía, AGE and JA (1999), in which the concentration of suspended solids was reported with a value of 26.87 g/L. A similar value of 30 g/L, recorded during the spill, was later presented (Ayala-Carcedo 2004; Ayora et al. 2001). However, this value was increased up to 660 g/L accounting for the estimated mass of muds deposited and the estimated spilled volume (Gallart et al. 1999; Martín-Peinado 2002).

The previous concentrations can be correlated with a very fine sediment particle size. CMA (1998) roughly estimated this parameter as less than 200 µm. Gallart et al. (1999) present two approaches: a mean particle size of 12 µm, ranging from 6 to 13 µm (ITGME 1998); and similar results from a laser beam scattering. A similar range, from 4.5 to 12 µm of the median particle-size (d50), was also presented (Querol et al. 1998; López-Pamo et al. 1999; Manzano et al. 2000; Vidal et al. 1999), while CSIC (2008) considers a higher range of 10 to 20 µm. These values agree with the d50 of the particles still stored in the pond: the pyritic lagoon (southern) had a d50 between 10 and 15 µm; but the pyroclastic lagoon (northern) had a wider range, from 18 to 250 µm (Alonso and Gens 2006b).

The bulk density was first estimated using a value of 3000 kg/m3 (CMA 1998). Other values are also found, such as 2850 kg/m3 (Penman et al. 2001) and 2950 kg/m3 (Martí et al. 2021). In the trilogy of papers published by Alonso and Gens (Alonso and Gens 2006a, b; Gens and Alonso 2006) the estimated density value of the liquefied tailings was 3100 kg/m3. Ayala-Carcedo (2004) used a fluid density value of 3100 kg/m3 for the muds and 2000–3100 kg/m3 for the acid water, with lower values after the disaster due to liquefaction and sedimentation processes.

Breaking Type and Time

The process of the breach formation, its shape, and dimensions are still unknown, as well as the main directions of the flow during its initial stage. Despite no direct measurements nor attempts to estimate them from the available information in the literature, it can be inferred that the breach formation was quasi-instantaneous (Alonso and Gens 2006a, b; Gens and Alonso 2006), and the fluid began to spill in large quantities immediately.

The breaking time was first presented in the official report of Organismo Autónomo de Parques Nacionales, OAPN (1998). The described time is 3:30 a.m., 10 min later than the neighbour’s notification (AGE and JA 1999). Despite the inconsistency in the breaking time, this time (3:30 a.m.) was preferred by some authors (Gallart et al. 1999; Grimalt et al. 1999; Moreno-Millán and Valdés-Morillo 2003). On the other hand, an earlier time between 0:30 and 1:00 a.m. is also extensively supported in the literature (Alonso and Gens 2000; Arenas et al. 2001; Ayala-Carcedo 2004; CSIC 2008; Fajardo et al. 2016; McDermott and Sibley 2000; Padilla et al. 2016).

Spilled Volume

Values of the spilled volume found in the literature range from 0.45 hm3 (Dorronsoro 2009) to 10.32 hm3 (Padilla et al. 2016). Frequently reported values of the spilled volume are between 4 and 6 hm3. Supplementary Fig. S-1 summarizes all these volumetric data (pond capacity and spilled volume).

It is important to remark that Moreno-Millán and Valdés-Morillo (2003) probably confused the type of fluids as the values they present are switched with respect to those proposed by others authors. Additionally, some discrepancies (see supplemental Fig. S-1) have been also found in the values presented by the same group of authors (Aguilar et al. 2002, 2003, 2004a, 2007, 2008, 1999; b, c; Dorronsoro et al. 2002; Martín et al. 2004, 2007, 2008; Simón et al. 2001, 2002, 2005). Despite that, the first estimate of 4.5 hm3 presented by CMA (1998) is the most referenced value.

Only a handful of authors explicitly referenced the volume of muds (Alastuey et al. 1999; Antón-Pacheco et al. 2001; López-Pamo et al. 1999), while some of them considered the whole spill to be a mud-like fluid (Benito et al. 1999; Guerrero et al. 2008; Hernández et al. 1999; Madejón et al. 2018a). López-Pamo et al. (1999) carried out a detailed cartography of the extension of the deposited muds, calculating their volume as 1.98 hm3 but omitting the volume of acid waters. In Fig. S-1 (green bar), the volume of the acid water inferred from the total and muds volumes is presented.

Spill Wave Propagation

The evolution of the flood was first said to be “very fast” for the first few kilometres and then decreasing to 1 km/h (CMA 1998). This value was used as the average velocity for the whole flood process by Garcia-Guinea et al. (1998). A general evolution of the flood is described in OAPN (1998), indicating an approximate arrival time at some locations (Fig. 2a). Other authors offer a brief description of the flood, probably based on the previous document, but omit some of the locations (Gallart et al. 1999; Martín-Peinado 2002).

The recorded data of gauge station EA90 was first introduced by Borja et al. (2001) (Fig. 2b, continuous line), showing a similar limnograph to that of the unpublished document of Consultec Ingenieros (1999) (Fig. 2b, dotted line). Both documents remark that it was impossible to obtain direct observations for high water depths, because they were outside the measurement range of the facility (limited to 2.5 m). According to Borja et al. (2001), the value of the first peak was 3.94 m, while the value used by Consultec Ingenieros (1999) of 3.86 m. In contrast, Alonso et al. (2010) suggest that the maximum water depth was 3.6 m.

Affected Length

The affected length of the spill also varies widely from 24 km (Alonso and Gens 2006a, b) up to 84.6 km (AGE and JA 1999), refereeing the lowest one only to the Guadiamar River. Supplemental Fig. S-2 summarizes the values of the total affected length found in the literature.

There is a scattered distribution of the affected length data, 40 km being the most referenced value (Aguilar et al. 2004c, 2007a; Alastuey et al. 1999; Antón-Pacheco et al. 1999, 2001; Cabrera et al. 1999; Gibert et al. 2011; Madejón et al. 2018b; Martín et al. 2008; McDermott and Sibley 2000; Moreno-Millán and Valdés-Morillo 2003; Palanques et al. 1999; Sánchez-Marañón et al. 2015; Santofimia et al. 2011, 2013; Simón et al. 1999; van Geen et al. 1999; WWF/Adena 2002). They probably assumed the first approximation presented by CMA (1998) to be correct. Other authors estimated a length between 60 and 63 km (Alonso et al. 2010; Arenas et al. 2001, 2008; Borja et al. 2001; Cabrera 2000; CHG 2005; CSIC 2008; Guerrero et al. 2008; Grimalt et al. 1999; Hudson-Edwards et al. 2003; WWF/Adena 2008).

In terms of affected length by the spilled muds, the detailed evaluation presented by López-Pamo et al. (1999) report a value of 45 km. However, other values in the literature range from 24 km (Alonso and Gens 2006b) to 60 km (Alonso et al. 2010; Guerrero et al. 2008), the most cited one being 40 km (Antón-Pacheco et al. 1999, 2001; Arenas et al. 2001, Ayala-Carcedo 2004; CSIC 2008; Dorronsoro 2009; Grimalt et al. 1999, 2008; Hudson-Edwards et al. 2003; Martín-Peinado 2002; Simón et al. 1998, 2007).

Regarding the acid waters spill length dimension, only Martín-Peinado (2002) described it explicitly with a value of 18 km. The values from other authors, depicted in Fig. S-2 (green bar), range from 10 to 23 km, and were calculated here as the difference between the total and the mud-affected lengths.

Affected Area and Deposited Muds

The total extent of the spill ranges from 2000 ha (Murillo et al. 1999) to 6000 ha (Alzaga et al. 1999). Supplemental Fig. S-3 depicts the values recompiled from the literature. CMA (1998) first defined the affected area at about 4400 ha. Then, this value was fine-tuned to 4634 ha (AGE and JA 1999). The most referenced value is 5500 ha (Aguilar et al. 2004a, c, 2007a, b; Martín et al. 2008; Simón et al. 1998, 1999, 2002). However, the same group of authors presented different values in other publications (Aguilar et al. 2002, 2003, 2004b; Dorronsoro 2009; Dorronsoro et al. 2002; Martín et al. 2004; Simón et al. 2001, 2008).

The area affected by the muds was analysed in detail by López-Pamo et al. (1999), indicating a value of 2616 ha. This is the most referenced value for the muds extent (Antón-Pacheco et al. 1999, 2001; Arenas et al. 2001; Ayala-Carcedo 2004; Benito-Calvo et al. 2001; Benito et al. 2001; Turner et al. 2002; WWF/Adena 2002, 2008), and it is not far from the 2703 ha initially presented by AGE and JA (1999). However, other authors reported a value of 2710 ha (Cabrera 2000; Cabrera et al. 2008; Martín-Peinado 2002), which is closer to the one in the official report (CMA 1998), or a higher one of 2886 ha that includes non-flooded areas inside the mud extent (Guerrero et al. 2008). On one hand, Grimalt et al. (1999) and Gibert et al. (2011), and on the other hand Pain et al. (1998), preferred the greatest values for the mud extent, which is 4286 and 5200 ha respectively. In these cases, the spill was probably all grouped as “muds”, including the acid waters.

The estimated width and thickness of the deposited muds also vary. On one hand, the affected width was first established to be 200 m CMA (1998), and later on corrected to 1000 m by CSIC (2008). Regarding the thickness of the deposited muds, some authors suggested high values near the pond, as for example 4 m by Turner et al. (2002), 3 m by CSIC (2008), 2 m (Gallart et al. 1999; Fajardo et al. 2016) or 1.7 m (Cabrera 2000; Grimalt et al. 1999; Hudson-Edwards et al. 2003). According the detailed analysis of López-Pamo et al. (1999), the thickness of the deposited muds was highly variable, ranging from less than 0.02 m to up to 2 m.

On the other hand, the area affected by the acid waters was stated early on to be 2000 ha (Pain et al. 1998) and later 900 ha (McDermott and Sibley 2000). These values differ from the official source (AGE and JA 1999), which reported a value of 1931 ha. For the other published values, the areas have been inferred from the total and muds areas (Fig. S-3, green bar).

Recovered Volume

A total volume of muds and terrain of 7 hm3 were reported to have been recovered after the spill incident and then deposited into another mine-pit (Antón-Pacheco et al. 2001; Arenas et al. 2001; Cabrera et al. 2008; CSIC 2008; Santofimia et al. 2011, 2013; WWF/Adena 2002). Arenas et al. (2001) and Cabrera et al. (2008) raised this value by 0.8 hm3, which corresponds to the second phase of the clean-up activities. Other values of the total recovered volume exist in the literature, such as 4.7 hm3 (Hudson-Edwards et al. 2003; Turner et al. 2002), 10 hm3 (Turner et al. 2008), or most recently, 8 hm3 (Madejón et al. 2018b).

According to Santofimia et al. (2011), who analysed in depth the different liquid dumps tipped into the mine-pit during the period 1995–2006, about 2.9 hm3 of recovered material was added during the time from the disaster to the end of the mining activities in 2001. Although the volume recovered during the summer of 2000 at the Entremuros area is not specified, it probably did not reach 1 hm3 (Arenas et al. 2001; Cabrera et al. 2008).

New Evidence

Construction and Filling Process

Two different data sources are presented and analysed here. Historical orthophotographic images of the Spanish Geographical Institute (IGN 2021) enabled an observation of a schematic evolution of the pond. In addition, topographical data were obtained from a few days after the incident by restitution of an ad hoc photogrammetric flight at a scale of 1:2.000 (Consultec Ingenieros 1999).

Figure 4a shows the area where the tailings pond was built, over a very light forest (image from before 1978). Figure 4b and Fig. 4c show intermediate construction stages (between 1980 and 1986), in which three lagoons are clearly seen: two larger ones, northern and southern, and one smaller (rectangular shaped) between them. Later, the intermediate lagoon was probably filled; only two lagoons remained just before the failure (Fig. 4d).

Fig. 4
figure 4

Pond characteristics and evolution of the construction process. Aerial photography taken by IGN: a Previous to 1978 (source: IGN, ‘Interministerial’ flight 1973–1986); b Intermediate stage I (source: IGN, ‘Nacional’ flight, 1980–1986); c Intermediate stage II (source: IGN, ‘Nacional’ flight, 1980–1986); and d six months before the pond-break (source: IGN, ‘Olistat’ flight, October 1997). e Map of the pond elevation after the pond-failure (source: adapted from Consultec Ingenieros (1999))

Full size image

The northwest part of the pond was the natural terrain slope and it seems, as shown in Fig. 4b, c, that an irregular surface defined its bottom (Fig. 1b). Therefore, it can be hypothesised that the pond bottom was not dug too deep, and the natural terrain could be used as the pond bottom. In early 1998, after stage 16 was completed, the east dyke almost reached the riverbed (Fig. 4d). The increment of the crest width was estimated at 35 m by comparing Fig. 4c, d, on the latest image.

Figure 4e illustrates the post-failure topography of the pond, where the elevation of the dykes was above 69 m at almost the whole crest-length (except for the area of the dyke failure). The lowest post-failure elevation of the pond was about 48 m; thus, the pond could have had a minimum height of about 21 m. Additionally, the topographical data shows that the external slope of the pond, near the Agrio River (east dyke, not affected by the breach), was approximately 1.2H:1V.

The area occupied by the fluid tailings during the construction process can be estimated using Geographical Information Systems (GIS) and photointerpretation techniques. At the intermediate stage (Fig. 4b), the fluid was irregularly distributed occupying about 78.9 ha. At the stage shown in Fig. 4c, the flooded area was about 105 ha. Finally, for the last image available (Fig. 4d), taken six months before the pond broke (October 1997), the pond was almost full, and the occupied area was about 140.8 ha. In this last stage, the northern lagoon occupied 105.6 ha, whereas the southern one was only 35.2 ha. In this sense, considering the internal perimeter of each lagoon, where the fluid reaches the top elevation of the dykes, and using the latest aerial photography as reference (Fig. 4d), the total area occupied by the fluid could have reached 155 ha (N: 115 ha; S: 40 ha), also far from the values generally observed in the literature.

The storage capacity has been estimated using historical topographic data from 1954 (Supplemental Fig. S-4) and the post-failure topography (Fig. 4e). The pond was located at the southwest side of the Los Frailes and Agrio junction, where the alluvial valley widens and the bankfull of the Agrio River did not reach 50 m. Considering that the bottom of the pond was not dug deeper, the lowest elevation of the pond was estimated at ≈48 m (Fig. 4e). Thus, assuming a flat bottom geometry with this elevation, a homogeneous crest elevation at 70 m, the dyke geometry shown in Fig. 1b, and a fluid height at 21 m, we determined that the pond’s calculated capacity was about 28 hm3 of tailings. Given all of these considerations, the exact real storage capacity just before the pond-failure, though still very uncertain, exceeded 20 hm3 and was most probably less than 28 hm3.

Fluid Type

It is worth noticing that the sediment concentration and density values range widely in the literature, and probably also varied along the river, decreasing downstream to very low sediment concentrations. This would have modified the fluid properties and, thus, its dynamic and static behaviour.

Videos recorded during the incident were analysed to provide a qualitative analysis of the fluid behaviour (Appendix B, Supplemental Table S-1). Although observations of the event are susceptible to interpretation, and the data acquired ad hoc during and after the event are only representative of particular locations, the available information is useful for understanding what kind of fluid or fluids were spilled (Sanz-Ramos et al. 2021b).

The criterion used to classify the fluid is based on an analysis of the observed flow patterns that are, or not, characteristic of the behaviour of a Newtonian fluid-type. In river engineering it is widely accepted that water, in its natural state, behaves as a Newtonian fluid. Thus, three kinds of fluid behaviours have been considered. For ‘water flow’, associated with Newtonian behaviour, two types of fluids are distinguished: A, or ‘clean water’, when the observed fluid seems like water without or with very little sediment; and B, or ‘not-clean water’, for water fluid-types with some concentrations of sediments, such as in natural rivers. ‘Hyperconcentrated flow’, denoted as C, is used for highly concentrated sediment-laden flow, i.e. an intermediate state between a Newtonian and non-Newtonian fluid (Beverage and Culbertson 1964; Costa 1998; Nemec 2009). Finally, ‘mud flow’, denoted as D, refers to a fluid in a dynamic or static situation that behaves like a non-Newtonian fluid. Depending on the registered image, the behaviour-type of the fluid may be a single type (e.g. “A”, “B”, etc.) or a combination of different behaviour-types (e.g. “B/C”, “C/D”, etc.).

Table 1 summarizes the number of times each behaviour-type of fluid has been observed in the videos (additional information can be found in Appendix B). The absence of observations of ‘clear water’ (A) contrasts with the high number of observations of ‘sediment-laden fluid’ (C), with more than 116 observations (isolated or in combination with other kinds of fluid behaviour). In general, when the fluid was flowing, the behaviour-type observed was “B” or “C”, besides their combination (B/C); while when the fluid was retained and deposited it seemed to be “C”, and/or “D”, and to a lesser extent “B”.

Table 1 Summary of the analysed videos. Number of times that a behaviour-type of the fluid, or a combination of them, is observed in the videos
Full size table

It is important to highlight that, a few hours after the pond failure, the fluid still spilling from the pond seemed more a hyperconcentrated fluid (C) than a mud-like fluid (D). This fact suggests that the core of the flood behaved more like a highly concentrated sediment-laden flow than like a mud flow, probably due to the fluidification of the mine-tailings initially sedimented and consolidated into the pond (Ayala-Carcedo 2004). The high sediment concentration of the fluid, as well as the sediment characteristics, mainly composed of very fine particles, could have favoured sedimentation processes generating the observed huge extensions of mud-like fluid.

Breaking Type and Time

The first aerial images taken after the spill (Fig. 5a) show how the pond was partially rebuilt immediately after the disaster (Fig. 5b). The zone where the three original dykes join suffered several modifications. However, a cliff can still be discerned at the east dyke of the northern lagoon, suggesting a breach at least 55 m wide (Fig. 5b, highlighted in green).

Fig. 5
figure 5

Description of the break type. a Aerial photography of the pond five days after the pond-failure (source: Junta de Andalucía, JA (2003)). b Zoom to the junction of the three dykes, the place where the breach was generated. c Sketch of a reasonable breach formation and evolution of the flooded area, the affected dykes coloured red

Full size image

The southeast dyke probably had a fan displacement with a rotation point about 500 m south of the east dyke’s junction. This type of failure also suggests that the main direction of the spill was to the north, at least in the first moments after the pond failure. This is in accordance with the flooded area observed upstream of the breaking point, more than 500 m upstream of the northern part of the pond (Fig. 5a). According to the topographic data of Consultec Ingenieros (1999), the flooded area reached the riverbed more than 10 m above the breakage. Only a breach that would allow the spill to flow against the natural slope could have flooded the riverbed and the riverbanks located 500 m upstream of the breaking point. After this upstream flow, the flow direction shifted to the east by eroding the dykes. Figure 5c presents a hypothetical sketch of the evolution at the first moment of the breach formation and the flood process.

The breaking time was probably first hydraulically estimated in the unpublished document of Consultec Ingenieros (1999). An inverse convolution process was performed to obtain an approximate hydrograph at the breakage point based on the EA90 gauge station information (Fig. 2a, location #1). This process was carried out by means of the numerical tool HEC-1. Ad hoc topographic data (Fig. 4e), which considers the deposited muds, was used to obtain the discharge–volume relations by means of HEC-RAS numerical tool. Finally, an iterative process was performed using the “Modified Puls” routing method to obtain the hydrograph and, thus, the breaking time. Despite the hypotheses used, Newtonian fluid (water) and the method of transforming the limnograph into a hydrograph, the results can be accepted as suitable enough, ranging the breaking time from 0:30 to 1:00 a.m.

Spilled Volume

The limnograph registered at EA90 gauge station (Fig. 2b) can be used to estimate the spilled volume using the rating curve of the gauge station to transform the water elevation into discharges. In this sense, the hydrograph at EA90 (Fig. 6a, dot line) was first presented in the unpublished document of Consultec Ingenieros (1999), with a maximum discharge of 600 m3/s. Borja et al. (2001) used the rating curve from Benito et al. (2001) to obtain the hidrograph with a peak discharge of 1056 m3/s. Later, using data from Palancar (2001), Ayala-Carcedo (2004) showed a similar graph, but considering the fluid to be water and muds (Fig. 6a, dashed line), and indicated that the estimated peak discharge of 811 m3/s was probably underestimated due to the higher viscosity of the fluid.

Fig. 6
figure 6

Data from the El Guijo gauge station (AE90). a Comparison of the hydrographs presented by Ayala-Carcedo (2004) (dashed line), by Consultec Ingenieros (1999) (dot line), and calculated from data of Borja et al. (2001) (continuous line) (*)using the rating curve from Benito et al. (2001). b Rating curve (**)obtained by numerical modelling (source: adapted from Benito et al. (2001)). c Estimated volume that passed through EA90 during the event (25–26 April), the white bar obtained being the mean value of the rating curve extrapolated from polynomial adjustment of the 3rd, 4th and 6th order. d Elevation-volume relative curve of the pond inferred from the post-failure topography

Full size image

Three rating curves have been inferred from the previous data. Figure 6b shows the best adjustment (R2 > 0.95) of the rating curve calculated here from the data of Consultec Ingenieros (1999) (dotted line) and Ayala-Carcedo (2004) (dashed line), and also the rating curve inferred from the data of Borja et al. (2001). These rating curves behave similarly for elevations up to 34.5 m, being more conservative the rating curve of Benito et al. (2001) (Fig. 6b, continuous line).

In this regard, data coming from the CHG, till now unpublished, related the mean daily water depth and discharges during 1998 at the gauge station EA90 located in the Guadiamar River (these data are detailed in “Discussion”). The rating curve inferred from these data shows a limited application range, up to 1.1 m, and there is no discharge value associated with the water depth of 25 April. For this reason, three different extrapolation methods, which consisted of a polynomial adjustment of the 3rd, 4th, and 6th order with a good adjustment (R2 > 0.99), were used to obtain the mean daily discharge corresponding to the water depth on 25 April, of 1.41 m.

Figure 6c shows the cumulative volume that passed through EA90 during the event, considering the three previously mentioned sources (grey-dashed bars) and new data presented here (white bar). These values were calculated by integrating the hydrograph data shown in Fig. 6a, and averaging the mean daily discharge calculated from the extrapolation methods for the CHG data. The volume ranges from 7.6 (Consultec Ingenieros 1999) to 14.4 hm3 (Borja et al. 2001), all of which exceed the volume usually referenced in the literature (4.5 hm3). Considering the uncertainties of the inferred rating curves for the hydrographs, the most reliable spilled volume could be 11.5 hm3, calculated from the average value of the mean daily discharge.

The main question is whether the pond had this much fluid stored before its failure. The initial and the final volumes in the pond are unknown, but the topographical data presented in Fig. 4e can be useful to assess the volume potentially spilled from the pond. Although the pond geometry was changed by the disaster, it is sufficient for this purpose because the modifications were slight (only the east dyke).

Figure 6d shows the elevation–volume curve obtained from the post-failure topography. The top elevation is revealed to exceed 70 m in several places by the post-failure topography (see Fig. 4e); thus, this value being the pond crest for the whole pond perimeter, the spilled volume could potentially reach 15.4 hm3. Nevertheless, according to Alonso and Gens (2006a), the water level was 4.4 m below the crest during the day before the failure; thus, the maximum spilled volume would be 9.3 hm3. Both presented and calculated values are within the volume that was potentially stored in the pond.

Finally, it is important to highlight that few authors have attempted to estimate the discharge hydrograph spilled from the pond. As shown in Figs. 2b and 6a, the registered flood was bimodal. The first peak was at about 3:30 a.m. and the second at about 9:00 a.m. Then, ≈ 48 h later, discharges dropped to values like those before the disaster. Even though the water depth overtopped the maximum calibrated value for the gauge station, the limnograph registers were continuous (Fig. 2b, dashed line).

In Consultec Ingenieros (1999), the hydrograph of the spill was estimated to be continuous by using the inverse convolution process mentioned above (Fig. 7a, dot line), with a peak discharge of 1050 m3/s at about 1:25 a.m. Most recently, Padilla et al. (2016) presented a hypothetical hydrograph, also with two peaks (Fig. 7a, continuous line). In contrast to the previous one, the second peak is higher than the first one, being about 1670 m3/s (at 1:30 a.m.) and 2500 m3/s (at 7:15 a.m.), respectively. In addition, these peaks were separated by a no-discharge period of 3:45 h, contrary to the observed limnograph (Fig. 2b, dashed line). The volumes of these hypothetical spills are 12.3 and 11.1 hm3, respectively (Fig. 7b), both being within the potential storage capacity estimated here.

Fig. 7
figure 7

a Hypothetical hydrograph of the spill at the failure point estimated by Consultec Ingenieros (Consultec Ingenieros 1999) (dot line) and Padilla et al. (Padilla et al. 2016) (continuous line). b Spill volume calculated from the previous data for the first 24 h

Full size image

In this sense, the volume calculated from the data of Consultec Ingenieros (1999) at the pond is 64% higher than that estimated at EA90 gauge station, suggesting that the rating curve used to transform the limnograph into a hydrograph was inadequate. On the other hand, it is unclear whether the spill was released in two time-separated phases as suggested by Padilla et al. (2016) because the dyke failure was very quick and probably affected both dykes concurrently (Alonso and Gens 2006a, b; Ayala-Carcedo 2004).

Spill Wave Propagation

The distances between the observation points (Fig. 2a) were obtained using GIS techniques and the ‘Olistat’ aerial photograph (IGN 2021) as a reference image (Fig. 4d). In addition, the breakage time has been established at 1:00 AM. These considerations have allowed us to properly estimate and compare the propagation velocity of the spill from the different authors analysed in this section.

Figure 2b illustrates the real flooding event monitored on a water depth trend curve at AE90 (dashed line), the arrival time being at 2:30 AM. Thus, considering the breaking time at 1:00 a.m., the maximum flood front velocity resulted in 4.7 km/h for the first stream. It is worth noticing that the first peak plots below 2.5 m because this value was the maximum one measurable using the limnograph. Later on, the facility operator wrote by hand on the limnograph that it was of 3.86 m, estimating it from the observed marks of the maximum flood elevation in a section of the gauge station.

Table 2 summarises the results of the wave front velocity calculated using the reference values of the literature and considering the corrected distances downstream from the incident location. The wave front velocity decreased along the river, which ends at Los Vaqueros ford (Fig. 2a, location #7). A double estimation was performed by Gallart et al. (1999): one considering the arrival time at two specific locations (Table 2, left column); the other based on the analysis of the two main slopes of the reach (Table 2, right column). Ayala-Carcedo (2004) also presents two different sets of values for the flood velocity: on one hand, referencing to Gallart et al. (1999) (Table 2, left column); and, on the other hand, assessing it with travel time data for just the first stream (Table 2, right column), till EA90 (Fig. 2a, location #1).

Table 2 Comparison of the front velocity of the flood (in km/h)
Full size table

There are too many uncertainties to establish the flood propagation velocity with precision. Particularly downstream of EA90 gauge station, observation points should be considered only as approximations because no data were recorded. Despite all that, the velocity of the flood front plausibly was slightly higher than 4.7 km/h from the breaking point to the EA90 gauge station since the breach first opened to the north (Fig. 5c) and, thus, the flood front should travel a larger distance in same time.

Affected Length

The affected length has been assessed by photointerpretation of the aerial images taken five days after the pond-failure, which correspond to a 1:10,000-scale flight of the Consejería de Medio Ambiente of Junta de Andalucía (JA 2003). All following distances along the Agrio and Guadiamar riverbeds were measured with GIS techniques and are represented as planimetric measurements.

The length of the affected streams initially considered to have been affected was 78.7 km, but adding the area upstream of the breaking point and the backwater zones in the Guadiamar River and Brazo de la Torre (Fig. 3d), the total affected length to be considered is 86.7 km. This agrees with the 84.6 km mentioned in the official report of AGE and JA (1999). This small difference is probably due to the measurement techniques used.

Particularly for the muds, they clearly stopped between the Don Simón and Los Varqueros fords because the colour of the spill alters from dark blue to orange (Fig. 3c). The distance affected by the muds, calculated by considering the deposits from the upstream area of the breaking point and the backwaters from Guadiamar–Agrio junction (Fig. 3b), would be about 52 km. This value is far from the estimations of López-Pamo et al. (1999), of 45 km, which is similar to the distance estimated herein till Don Simón ford (see Table 2), of 44.6 km. As the transition area is difficult to determine exactly, the affected length by the muds would range from 44.6 to 52 km.

The Entremuros area retained about 2–3 hm3 of acid waters. It is worth noticing that on the days after the spill it rained (see “Discussion”) and, according to OAPN (1998), discharges from adjacent agricultural fields could have raised this volume up to 5 hm3. To confirm this, the storage capacity of Entremuros has been evaluated using recent Digital Terrain Elevation (DTM) data, being a good approximation because this area did not experience severe changes after the pond-break, even after the restoration activities (AGE and JA 1999; Arenas et al. 2008).

Table 3 captures the extents of the acid water for different volumes retained in Entremuros, and their corresponding water elevation. Small changes in the water elevation generate huge increments in the retained waters. The flood extension in the lower part of Entremuros would reach a few meters upstream of the Don Simón ford (Fig. 3c), the length affected by the acid waters staying about 21.5 km for all volumes assessed.

Table 3 Relationship between stored volume (V), water elevation (Z), and flooded area (A) in Entremuros
Full size table

In addition, images taken 5 days after the spill at the lower part of the affected area suggest that the fluid reached the Guadalquivir River (Fig. 3d). This is in accordance with OAPN (1998) and Grimalt et al. (1999), suggesting that the retained fluid in Entremuros overtopped the partially built dyke at El Cangrejo Grande (Fig. 2a, location #9). Afterwards, the fluid reached the Guadalquivir River and potentially its estuary (Palanques et al. 1999). Considering this situation, the length affected by the acid waters could increase by up to 36 km.

However, CSIC (2008) indicates that the effects of this small spill were negligible. In this sense, Grimalt et al. (1999) and Cabrera et al. (1987) remark that contaminants from mine effluent leaks were already deposited at the Guadalquivir estuary before the disaster. Thus, data inferred by polluted soils in this area should be used carefully to estimate the ultimate affected length of the acid waters and, thus, the total affected length.

Affected Area and Deposited Muds

In terms of the total affected area of the spill, applying the above-mentioned photointerpretation techniques, we calculated the same value as AGE and JA (1999): 4634 ha. The area affected by the muds is 2703 ha, exceeding the 1931 ha affected by the acid waters. However, the mixing area is not clear, as previously shown.

The widest affected section, calculated here by tracing sections perpendicular to the riverbed direction, was about 2000 m at the Don Simón ford (Fig. 2, location #6). However, this value is at the site where the Brazo de la Torre enters the Entremuros area and where, additionally, the Guadiamar River branches into the old and new riverbeds (Fig. 3c). Considering only the main streams, the widest section affected by the flood was about 1400 m.

Recovered Volume

A detailed description of the recovered volume for each allotted area is still unknown. Interactions with the environment, such as basin inputs and outputs during the recovery, could have increased or decreased the estimated final recovered volume (see “Discussion”). In any case, it is expected to be between the lowest reported value of 1.98 hm3 (López-Pamo et al. 1999) and the greatest of 9.9 hm3, considering 7 hm3 from the recovering activities and 2.9 hm3 from the mining activity.

According to the total flood extent and mud volume described by López-Pamo et al. (1999), an average terrain thickness of 0.04 m was removed. This value is increased to 0.07 m if only the area occupied by the muds is considered, which is within the values described in the literature (AGE and JA 1999; Alonso et al. 2001; Antón-Pacheco et al. 2001).

Discussion

Basin Inputs and Outputs

The shown variability of the initially stored, spilled, and finally recovered volumes described in the literature increases when hydrological processes are considered. An analysis of unpublished hydrological data from the CHG and additional meteorological data from the Spanish Meteorological Service (AEMET), can help to clarify some of the uncertainties about the volumetric data.

To that end, the hydrological basin that drains the overland flow into the affected area was considered (depicted in Supplemental Fig. S-5), which was monitored by seven hydro-meteorological facilities: two water-level gauges (EA56, located upstream of the study area, and EA90) and the rainfall gauge E67 from CHG; and four rainfall gauges coded as 5783 (Sevilla), 5860E (Moger-El Arenisillo), 5960 (Jerez de la Frontera) and 5910 (Rota) from the AEMET (2020).

Daily cumulative rainfall data (Supplemental Fig. S-6) reveal three different patterns during the 1998 April–September period. From April to May, the weather was quite wet with more than 30% rainy days. From May to mid-September, the precipitation in that area was scarce or zero. From mid- to late-September, rainfall events were registered again on about 50% of rainy days.

Rainfall before the disaster could have increased the water level in the pond. Considering the E67 rainfall gauge as representative of the rainfall over the pond, a gross rainfall of 23 mm accumulated from 1 April till the pond-break, but only 3 mm precipitated the week before the failure. Thus, considering a pond extend of 155 ha, an extra volume of about 4.65 hm3 could have been stored in the pond due to rainfall. A sudden increment in the stored volume could have triggered the disaster (Rico et al. 2008) by generating an extra load on the terrain and dykes.

The collected water-level gauge data have been grouped by days, and the mean values of the height and the registered discharge have been represented. Figure 8 shows the water level evolution for 24 weeks at each gauge station, starting a week before the failure. At the upper part of the Guadiamar River, gauge station EA56 (Fig. 8b) reports a mean daily discharge of about 0.191 m3/s, or 0.016 hm3/day, with an important increment of 2.76 hm3 during the rainfall event of the last week of May.

Fig. 8
figure 8

a Evolution of the mean daily height (blue dots) and the cumulated volume (grey bars) at the gauge station EA56 (a) and EA90 (b)

Full size image

The EA90 gauge station could have not correctly registered the peak discharge of the spill during the event. However, according to the presented rating curve obtained from the CHG data, based on the mean daily height of 1.41 m registered on 25 April, the cumulative volume that passed through EA90 was estimated at 11.5 hm3 (Fig. 8a). However, a sharper increment of the accumulated volume of about 1 hm3 was registered at EA90 during 27 to 28 May (Fig. 8a). This value does not agree with the runoff generated in the subbasin lateral to Guadiamar River between EA56 to EA90, which was registered as 0.096 hm3 at EA56 (Fig. 8b). During the previous days, the rainfall in the area, especially in the closest rainfall station (E67), was below the average initial abstraction, about 20–25 mm (Fig. S-5). Thus, the volume increment registered at EA90 cannot be justified based on only the input coming from upper Guadiamar River (Fig. 8b). Discharges from the Agrio Reservoir, located upstream of the pond in the Agrio River, could plausibly be the cause of this anomalous data; however, no data is available.

One of the restoration activities during 1998 was the construction of a sewage treatment plant. First, a mobile emergency plant was installed on 1 July (Ayala-Carcedo 2004), but it did not start working till 24 July (Arenas et al. 2001, 2008). This plant worked till 21 August and treated 1.64 hm3 (Arenas et al. 2001). Later on, a new facility was built and began to work till 29 September and treated 1.15 hm3. Thus, the minimum amount of spilled acid waters and surface runoff retained in Entremuros based on the treated water volumes sums up to 2.79 hm3.

During those days of 1998 (25/4–29/9), the natural runoff flowed into the affected area, plausibly passed over the deposited muds without infiltrating, and thus converted the runoff into polluted water. According to the CHG data, until 29 September, the study area accumulated at least 5.4 hm3 of additional acid waters (no data from other inputs in the subbasin are available). This value agrees well with the 5 hm3 of acid waters retained at Entremuros (Arenas et al. 2001; CHG 2005; Grimalt et al. 1999) and with the retained potential capacity of fluid of the Entremuros area (AGE and JA 1999; OAPN 1998), which has been confirmed herein.

Additionally, Martín-Peinado (2002) and Grimalt et al. (1999) suggest that occasional spills from the agricultural fields entered into Entremuros, increasing the inputs and the retained water in this area. Although the amount of clean water discharged from the fields to the affected area is not described, according to the results presented herein, this area could have an additional storage capacity of more than 2 hm3.

Thus, less than half of the runoff was not treated. However, considering that 1.53 hm3 of clean water was directly spilled into the Guadalquivir River from 23 July to 2 September (AGE and JA 1999), the non-treated volume was reduced to 1.08 hm3. This volume, and the volume probably spilled from the fields, both infiltrated (Alcolea et al. 2001; Ayora et al. 2001; Manzano et al. 1999, 2000; Olías et al. 2006) and evaporated. In this sense, the intermediate period from May to mid-September probably favoured desiccation of the muds. Some pictures illustrate this desiccation process (Arenas et al. 2008; Cabrera et al. 2008; CSIC 2008), but neither the date nor the location is indicated.

The muds penetrated into the natural soil due to the difference of granulometry between the terrain and the muds (Alonso et al. 2001). This aspect could be magnified by the rainfall, particularly in the weeks after the disaster, infiltrating part of the deposited muds and polluting the natural soil and groundwater (AGE and JA 1999; Manzano et al. 1999).

Morphological Changes and Restoration Activities

A detailed description of the planned restoration activities is available in AGE and JA (1999), and those finally executed are reported in Arenas et al. (2008). Thus, only changes from the hydraulic point of view are discussed in this section.

As mentioned above, the restoration activities began with the clean-up of the muds and the treatment of the acid waters retained at Entremuros. According to AGE and JA (1999), two different strategies were applied for removal of the muds and the polluted terrain, depending on their location.

For the riverbeds and riverbanks, the works were carried out in sections. The river was first deviated, and sections were isolated by levees to create dry conditions. Where the extent of the riverbanks was excessively large, transversal platforms were built up to allow the heavy machinery access. Once the clean-up activities ended, the sections were again modified, aiming to achieve a morphological situation similar to the original. In the agricultural fields, the strategy followed was also using machinery to remove the muds and 0.05–0.1 m of superficial soil (AGE and JA 1999). A manual clean-up was executed to restore the locations where the machinery was not able to achieve satisfactory results. The Entremuros area, mainly affected by the acid waters, was cleaned-up by removing only 0.02–0.04 m of soil. Thus, the morphological changes in this last area were reduced.

Although these activities should follow strict specifications, the clean-up area allotted to the mining company was dug in excess because they used very heavy machinery, generating residual contamination problems on the ground (AGE and JA 1999). Thus, significant morphological changes were produced from the mine pond to Sanlúcar la Mayor (Antón-Pacheco et al. 2001), as shown in Supplemental Fig. S-7.

Herbaceous crops, plantations, bushes, and trees were also removed as part of the clean-up activities. The ‘Corredor Verde del Guadiamar’ and the ‘Doñana 2005’ projects attempted to re-naturalise not only the affected area but also nearby ones (Arenas et al. 2008; CHG 2005). According to CSIC (2008), implementation of the ‘Corredor Verde del Guadiamar’ considerably modified the morphology of the rivers. After a homogenization of the superficial layer with an average increment of 0.25 m, reforestation activities were promoted to restore the new morphology (Cabrera et al. 2008).

The main change was the incorporation of the affected agricultural areas into the riverbanks. The Las Doblas gravel pit located just upstream of the Sanlúcar la Mayor bridge (Fig. 2a, location #2), which was very degraded before the disaster (Arenas et al. 2008), is an example of these restoration measures and the changes in the river and riverbanks morphology (Fig. 9). The new Las Doblas wetland was reconnected to the river, and nowadays is an important recreation area of the ‘Corredor Verde del Guadiamar’.

Fig. 9
figure 9

a Evolution of the wetland of Las Doblas: a 1997, previous to the spill (source: IGN, ‘Olistat’ flight, October 1997); b 1998, five days after the spill (source: 1:10,000-scale flight from JA (2003)); c current state (source: IGN (2021), most recent orthophoto of the ‘PNOA project’)

Full size image

Flood Event Reproduction

The aforementioned morphological changes in the river and their riverbanks may condition any numerical reproduction of the flood event, especially its hydraulic behaviour (Sanz-Ramos et al. 2020). Considering that the topographical data available prior to the disaster was a 1:10,000 contour lines map and that the downstream area is extremely flat, some parts of the affected area could differ enormously from the currently available topographical data.

In this sense, only a handful of hydraulic studies have been carried out till now, probably due to the lack of topographical resolution, besides the discrepancies presented and discussed herein of the hydraulics of the event. Simulations accomplished by subsurface and surface regional hydrological modelling (Padilla et al. 2016) were based on a military topography and then discretised by a triangular mesh with an average density of 4.8 elements per ha, far from the values commonly used for hydraulic characterization (Sanz-Ramos et al. 2021a). In contrast, and probably looking for more spatial resolution, the numerical modelling presented by Castro-Díaz et al. (2008) uses the post-failure topographical data of Consultec Ingenieros (1999). The differences presented in the flood extension could be associated with the terrain geometry used, in which the deposited muds are considered. In both cases, the analysed area did not include the lower part of the area affected by the spill, and the fluid was considered to be of the Newtonian type (water).

Recently, simulations carried out by Sanz-Ramos et al. (2020) implemented a non-Newtonian approach based on a Bingham-type fluid for the first few kilometres of the flood. In this approximation, the results of the propagation of the spill from the breakage point agrees well with the observed hydrograph at gauge station EA90. However, the differences in the flood extent, particularly at the area close to the pond, could be due to the topography used and the hydrograph imposed as an inlet, which corresponded to the one calculated in the Consultec Ingenieros (1999).

It might be interesting to numerically simulate the full extent of the spill. The fact that no one has done so is probably due to the uncertainties in the real spilled volume and the kind of fluid spilled from the pond.

New Evidence on the Hydraulics

The topographical analysis presented herein based on the post-failure topographical data of Consultec Ingenieros (1999) shows that the maximum elevation of their dikes was up to 70 m and the lower part was 48 m. Thus, a maximum fluid height of 22 m could have been present in the pond before the dyke-break. Assuming a fluid height of 21 m and considering the available geometrical specification of the construction process planned, the maximum potential storage capacity just before failure would have been 26.5–28 hm3 of fluid and consolidated sediment.

The spilled volume was only registered at the El Guijo gauge station (EA90), located 7.1 km downstream of the breakage point. Although the gauge was overtopped during the flood, and some data could not be properly obtained (Consultec Ingenieros 1999), the higher of the two registered peak discharges of the flood probably exceeded 1000 m3/s (Consultec Ingenieros 1999; Sanz-Ramos et al. 2020). Previously unpublished data of the CHG of the mean daily discharge registered at EA90, related with mean water depths, provides enough data to extrapolate the rating curve of the gauge station. The mean value of the three different extrapolation methods used herein demonstrates that the spilled volume could have been about 11.5 hm3. This exceeds all of the volume values found in the literature (about 5–6 hm3), but agrees well with the maximum volume potentially spilled from the pond, which was 15.4 hm3 according to the post-failure topography of Consultec Ingenieros (1999). In this regard, according to McDermott and Sibley (2000), the volume that was retained in the pond after the failure was 13.7 hm3. Adding the spilled volume of 11.5 hm3 calculated herein, the total fluid stored in the pond was about 25.2 hm3, a value closer to the pond capacity just before the break (26.5–28 hm3).

The spill, which started about 0:30–1:00 a.m., was first registered at 2:30 a.m. at EA90. The fluid was rapidly released from the two lagoons at about the same time, probably to the north direction first, contrary to the natural flow direction, and then to a more natural direction as the dyke eroded (Fig. 5c). The breach evolution described in the “New evidence” section above agrees with the flooded area observed upstream of the breakage point.

The probable mixture of the two kinds of fluids after the pond failure, in addition to the fluid characterization carried out by the video interpretation (Appendix B, Supplemental Table S-1), reveals that the fluid was more a highly concentrated or hyperconcentrated sediment-laden flow, than a mud-like flow as generally denoted in the literature. The flow probably behaved intermediately between a Newtonian and non-Newtonian fluid. One of the consequences of the spill was the deposition of huge amounts of fine sediments, which certainly seemed to have a mud-like behaviour once deposited, though it seemed to have moved more like a water like-fluid with suspended sediment. The scarce studies carried out to estimate the suspended solid concentration during the spill present a very wide range of values, between 26.87 (AGE and JA 1999) and 660 g/L (Gallart et al. 1999).

The fluid properties condition the flood propagation. An almost complete description of the spill front position-time observations is described only by OAPN (1998). However, the locations of some points do not allow extraction of the front velocities of the flood with confidence. Only an average velocity about 4.7 km/h during the firsts 7.1 km can be inferred from the limnograph of EA90.

The re-analysis carried out reveals that the total length of the reaches affected by the spill was 86.7 km, which is only 2.1 km greater than the value reported by AGE and JA (1999). This includes the 5 km long reach upstream of the breaking point, and the 2.2 km and 3.9 km long reach flooded by the backwater effect at the Agrio–Guadiamar junction and the Brazo de la Torre, respectively. Considering the upstream distances from the pond and the backwater effect, the present re-analysis increases the length affected by the muds to 52 km.

The total affected area was clearly defined as 4634 ha by the official report of AGE and JA (1999), composed of 2703 ha of muds and 1931 ha of acid water. However, according to López-Pamo et al. (1999), the area affected by the muds was 2616 ha. These differences can be attributed to consideration of the non-flooded areas and the area where the muds and the acid waters were mixed, which cannot be clearly defined.

Mining Since the Aznalcóllar Disaster

The Aznalcóllar disaster generated a huge environmental damage to the area (CSIC 2008), besides a notable social alarm (Elías 2001). At that moment, it probably contributed to a worldwide general deterioration of the public acceptance of mining (Ayala-Carcedo 2004).

Guidelines for design, construction, and closure of tailing dams, and to manage and prevent hazardous situations, had been developed previous to the event, and following the recommendations of these guidelines can greatly reduce the risk of a failure or the occurrence of dangerous situations for people and the environment (Penman et al. 2001). Management and monitoring strategies for active and abandoned mines can also help (Fernandez-Rubio et al. 1987; Kheirkhah Gildeh et al. 2021; Morton 2021). These strategies are necessary and should be considered mandatory, even for mine reclamation activities after decommissioning (Klose 2007).

Despite that, 64 new major tailings dam failures have occurred since April 1998 (WISE 2020). Since mine disasters are reported almost twice a year, additional management, prevention, and planning strategies are needed. To that end, numerical models have become valuable tools for resolving many engineering problems dealing with mining activities, especially those with safety and environment interactions (natural or induced by hazardous situations). Knowing the tailings pond characteristics (construction and filling process) and the kind of fluid stored (preservation or not of Newtonian properties) is necessary because numerical tools need such data to assess geotechnical, hydrological, and hydraulic aspects with confidence (Alonso 2021; Castro-Díaz et al. 2008; Kheirkhah Gildeh et al. 2021; Kuchovský et al. 2017; Martí et al. 2021; Padilla et al. 2016; Sanz-Ramos et al. 2020).

Additionally, ad hoc guidelines that allow the complexity of fluid and tailings dam failures to be considered must be developed to properly characterize potential breach and runout scenarios (Kheirkhah Gildeh et al. 2021). Such guidelines should encompass each particularity of any analysis of potential tailings dam breach and runout scenarios and should be adapted to the legislation of each country and aligned with the various monitoring strategies to control the construction and rehabilitation of tailings dams.

Conclusions

This paper compiles, discusses, and points out facts and uncertainties regarding the hydraulics of the Aznalcóllar mine disaster, such as the pond and fluid characteristics, the break type, the breaking time, the flow propagation, the volume potentially stored and subsequently spilled, area dimensions affected by the spill, and the morphological changes generated in the river and riverbanks caused by the spill and later by restoration activities. Aiming to clarify the most controversial data, new evidence was presented, analysed, and discussed from a hydraulic point of view.

The new evidence show that the spill could have had a volume of about 11.5 hm3, twice the most referenced values. In addition, hydrological evidence indicates additional inputs of water from the drainage basin, increasing the spill volume to more than 5 hm3 during the period from the disaster to the end of September 1998.

According to the video interpretation of the fluid’s behaviour, and in contrast with the common flow behaviour descriptions in the literature, the fluid flowed more like a highly concentrated sediment-laden flow than a mud-like flow. This agrees well with the relatively high values of the flood front velocity that have been estimated.

Finally, the restoration activities morphologically changed the affected rivers and their riverbanks. These changes may condition numerical reproduction of the flood event, especially its hydraulic behaviour, if simulations are performed with the topographical data obtained after the event. These morphological changes resulted in substantial modification of the hydro-sedimentary behaviour of the hydrographic network ending in the Doñana National Park marshes that still persist.