Establishing a Hydrographic Framework for Watershed Management across Northern Chile

This article demonstrates how currently available digital elevation (NASA SRTM; 30 m resolution) and hourly global precipitation data (rain, snow; 5 and 10 km resolution) can be used to hydrographically quantify the regional watershed management context across northern Chile. This is done through comprehensive derivations of flow direction, flow accumulation, flow channels, floodplain extent, depressions, and upslope watershed outlines. In turn, these derivations allow for estimating potential precipitation accumulations within any watershed, and turn these into subsequent storm-averaged discharge estimates at, e.g., at any road—flow-channel crossing points. This article elaborates on this by modelling and mapping hydrological conditions and subsequent storm damage at the regional scale and at select locations in terms of recent flood events on March 2015, May 2017, and June 2017. It was found that modelled flood extent and storm-estimated discharge volumes and rates generally conform to reported values including storm-caused damages within communities along the Huasco, Elqui, Limari, Copiapo and Salado rivers. This included the storm response assessment of six water reservoirs as these varied, as quantified, from normal (Puclaro, La Laguna, Cogoti), at capacity (La Paloma, Cogoli), and overflowing (Recoleta). The details of the local to regional assessments are presented in the form hydrologically explicit maps, figures and tables. Together, these attest to the general validity of the framework as introduced. Hydrometrically based stream-discharge calibrations would assist in further refining the approach, especially in terms of estimating local to regional run-off coefficients.


Introduction
Arid conditions combined with occasional heavy rainfall events are leading to catastrophic flashfloods, as have been reported for Northern Chile [1]. For example, on March 25-27, 2015, a 3-day rainfall, amounting to 75 mm, displaced about 30,000 people across the Antofagasta and Atacama regions through mudflow and flooding [2] (Figure 1). This included a rising death toll, more than 8,000 houses damaged or destroyed, extensive road blockages leading to community isolation, compromised water supplies, and temporary suspension of industrial activities. Similar rain events tend to be benign elsewhere. The generally parched and barren soil and bedrock terrain, however, is prone to transform otherwise ordinary rainfall events into valley-focused landslides, flashfloods, and mudflows [3]. An in-depth geophysical analyses of the March 25-27, 2015 flood event [4] [5] focussed on the Salado River watershed and related flash-induced mudflow impacts on the communities along this river, notably Copiapó, Diego de Almagro and Chañaral. It was found that the release of mud was mainly caused by floodplain re-channeling, coupled by flow-retarding debris that accumulated within dry ravines and riverbeds prior to the event [6]. Recent flood events across the Coquimbo province (May 12, 2017) and Antofagasta (June 8, 2017) were also damaging, but not to the same extent.
In reference to flood-related damages including mobilizations of contaminants, [14] and [15] commented on benefits, challenges, legal requirements and innovations regarding protecting waterways downslope from active and inactive mines. Primary options deal with dam stabilization, stacking dewatered mine tailings away from watercourses, and advancing towards contaminant-free metal extraction and refinement processes. Afforestation would also play important in terms of soil stabilization and enhancing rural developments [16].
This article focusses on quantifying the hydrographic flow channel, depres-sion and watershed context across northern Chile in a locally to regionally comprehensive manner. This was done by way of combining established digital elevation modelling routines with NASA's globally available precipitation distribution data in order to estimate amounts of water received and flowing across storm-impacted watersheds. In reference to storm-impacted accounts and analyses, emphasis is on building a comprehensive overview on where and how much water could potentially flow towards and through communities across northern Chile for any particular precipitation event.

Methods
The Northern Chile precipitation events for March 25-27, 2015 and May 12,   2017 were captured for the Arica, Tarapacá, Antofagasta, Atacama and Co-quimbo regions using NASA's georeferenced NEX-GDD Precipitation [17] and GPM [18] rain and snow data layers. These data were re-projected (utm 19s) and re-interpolated to match NASA's SRTM digital elevation coverage (30 m resolution) [19] across the regions. This data capture was extended beyond Chile's borders to ensure complete cross-border watershed coverages to assess the total amounts of precipitation received per watershed and per duration of storm event.
The re-projected digital elevation model (DEM) served to delineate local to regional flow networks using the D8 algorithm to determine flow direction and flow accumulation from the filled DEM [20]. Doing so established the flow direction rasters and the resulting pixel-specific flow-accumulation pattern.
Depressions were located by subtracting the non-filled DEM from the filled DEM. All depressions with significant water retention capacity up to their pour points were mapped by extent, depth, volume, and upslope watershed areas. The DEM-derived flow-channel networks were obtaining from the flow-accumulation raster by setting upstream flow-accumulation thresholds at 40, 400, 4000 and 40,000 ha, followed by raster to polyline conversion. This was done to emulate the extent to which the flow networks would advance towards or retract from higher elevations towards depressions and the coast by season and climate.
The extent of depression-terminated flow channels was emulated stepwise using increasing water-retaining volume capacity thresholds from 0.1 ML to 10 GL (1 ML = 10 3 m 3 ; 1 GL = 10 6 m 3 ) in order of magnitude steps. The capacity of the depressions to retain water was estimated by summing the 30 m × 30 m pixel depths across each depression. Flow channels were set to terminate at the lowest point in depression unless ready to overflow at the depression outlets.
The maximum amount of water potentially flowing along the DEM-derived flow channels was estimated by assigning the raster-based NASA-precipitation or rain amounts to each DEM raster cell, and summing these amounts along each flow accumulation path towards each water-receiving depression including the Pacific Ocean. The resulting sum was corrected by the amount of water retained within each upslope depression. The sum of water so accumulating was divided by storm duration to estimate the storm-average flow rates per event in m 3 ·sec −1 .
In principle, the numbers should serve as first-approximation location and event-specific flow-severity metrics. Hydrometric adjustments would be needed to account for, e.g., run-off diminishing evapotranspiration losses, snow sublimation, and soil and groundwater storage. Further adjustments may be required to account for actual amounts of water retained by upstream water reservoirs International Journal of Intelligence Science  Other datalayers (shapefiles) relevant for establishing hydrographic context were obtained from the internet, i.e., Open Street Map (OSM) for roads and streams [24], Map Cruzing [25] for places and landuse, DIVA-GIS [26] for administrative borders, roads, water and watercourse features, List of Saltpeter Works in Tarapacá and Antofagasta [27], and Mine Search [28]. Mining locations were also located based on latest available Google Earth satellite imagery (2014-2016) by placemarking each location while referring to open-pit and underground mines, ore-processing, and tailing ponds. The flowchart in Figure 3 presents the framework by which NASA-compiled precipitation layers were combined with STRM-generated DEM layers to determine 1) how much water was received within each DEM pixel, and 2) the extent to which these amounts accumulated along the DEM-determined depression and flow accumulation patterns. The resulting depression-overflow pattern was used to develop the regional to local flow networks. These varied by storm event from terminating in non-overflowing depressions (hence endorheic) or reaching the Pacific shore (hence exorheic). Open Street delineations were used to automatically 1) locate all road-stream intersection points and their upstream flow-retaining and -accumulating watershed areas, and 2) determine total storm flow volumes and storm-averaged discharge rates. The slope and flow-channel generated DTW derivations were used to delineate flood and soil moisture regime extent zonations above water-containing flow channels and water bodies. The entire framework was developed and coded for seamless ArcGIS processing, with special attention given to terminating flow channels in non-overflowing depressions. All input and output data layers were assembled in the form of an ArcMap package, and were used to produce the maps shown below. Figure 3. Hydrographic framework for watershed management for storm-based water flow, retention, and flood extent estimation using local to regional DEM-generated flow channel and depression delineations with emphasis on road-stream crossings. Table 1 lists major cities and towns where major flood damage may or may not have occurred for the March 2015 and the May and June 2017 storm events. Also specified are: 1) upslope basin areas (in km 2 ); 2) percent of floodplain coverage within upslope basins; 3) amount of precipitation (snow, rain) received per upslope basin as derived from NASA's precipitation rasters; 4) corresponding estimate for average flow rates per storm duration by locations, assuming no water retention or loss per storm duration.

Results and Discussion
March 2015 Flood Event, Northern Chile Presented in Figure 4 is an overview of the data layers used for establishing the hydrographic flash-flood context for the four northern regions in Chile pertaining to the 25-27 March 2015 precipitation (rain) event. Shown are: 1) the NASA captured precipitation pattern; 2) the DEM-derived delineation of watersheds receiving significant amounts of precipitation; 3) the SRTM-captured variation in elevation (hill-shaded) 4) the distribution of water-retaining     (Table 1).
Further north at Chanaral, at the mouth of Rio Salado (Figure 7, left), a similar debris-constrained and mud-filled flashflood caused catastrophic damage.
According to [4], flow rates peaked at about 1450 m 3 ·sec −1 , with 1150 m 3 ·sec −1 for water only. Table 1 estimates for the average water-only flow rate during the 51  [4]. For Diego de Almagro (Figure 7, left), mud-produced damage was also devastating, with an estimated peak flow rate at 310 m 3 ·sec −1 [4], while the storm-averaged water-only flow rate in Table 1 amounted to 75 m 3 ·sec −1 .
These large differences between peak and storm-averaged flow rates are undoubtedly due to flow-retarding debris at the head of the floods at these locations [6], and at Copiapó as well.   (Table 1). This, however, did not happen as could be inferred from the February 2001 flood analysis for Calama [31] [32] [33]. According to this analysis, much of the water Figure 12. DEM-delineated water-contributing watershed areas (hill-shaded) towards Antofagasta pertaining to the NASA captured June 8, 2017 storm event, assuming that only SRTM-DEM derived depressions ≥ 0.1 ML retain water. Left: precipitation, flow channels (white) and floodplains (brown; includes areas with DTW ≤ 40 m above flow channels when bankful), with storm-based flow amounts in GL at select locations. Right: elevation pattern with storm-specific flow rates in m 3 ·sec −1 (water only, mud excluded) averaged over the storm event. Due to persistently arid conditions, combined water retention and loss due to evapotranspiration and soil infiltration likely would have been at or near 100% along the south-to-north stretching watershed, but not so along the northern part, where stormwater flows downward along steep flow channels into the coastal floodplain. Yellow lines: Open Street roads. International Journal of Intelligence Science    Table 1). The reason that Calama did not flood in 2015 but did so in 2001 relates to the fact that the February 2001 storm event was centered on the Ata-International Journal of Intelligence Science cama desert, and this included Calama and the steep terrain north and northeast of the Chuquicamata open pit mines [33].
Of special note is the SRTM-DEM mapped depression above the Rio Loa gorge east of Calama (Figure 9). This DEM-generated depression is likely artificial due to geospatial 30 m grid-point interpolation across steep terrain and particularly so across the entrance point (pour point) of this gorge. Due to this artifact, the depression could have filled up to 86% of its estimated water-holding capacity. Inspecting the surface images provides evidence that water-filled and salt-enriched depressions exist in this area as mapped in Figure 9.  to completely closed to ship traffic. The estimated maximum average discharge rates set in Table 1 for rain-induced storm discharge duration without upslope water retention amounted to 2589 and 3565 m 3 ·sec −1 at La Serena and Ovalle, respectively. Accounting for upslope water retention dropped the flow rate estimates in Table 2 to 1752 and 2089 m 3 ·sec −1 , respectively. At Ovalle, however, flooding damage was extensive by reaching into residential areas as reported and mapped in Figure 11 Three of the water reservoirs within the Rio Limari, Rio Elqui and Rio Choapa watersheds are estimated to have accommodated their upstream inflow on May 12, 2017, with the Laguna Reservoir basin receiving only snow, as detailed in Table 2. Two of the reservoirs (La Paloma, Cogoti) were at capacity, and likely overflowed through spillway discharge. The Recoleta reservoir, however, experienced heavy overflows, leading to a partial collapse of its upper wall section (http://www.elmostrador.cl/noticias/multimedia/2017/05/12/video-el-colapso-del -embalse-recoleta-en-ovalle-que-obliga-a-evacuar-zonas-aledanas-al-rio-limari/).
June 2017 Flood Event, Antofagasta Region Journal of Geographic Information System Another severe rainfall occurred across the Antofagasta region on June 8, 2017. Details about the rainfall distribution of this event and the flow-channel pattern across the hill-shaded watershed areas upslope from Antofagasta are presented in Figure 12 (left). Also shown is an example of how average stream discharge amounts and rates would have varied across the upslope watershed areas assuming no water loss and retention other than what would fill all SRTM-DEM derived ≥ 0.1 ML depressions. No water would have reached the coast if all DEM depressions with as little as 1 L are also be assumed to be water retaining. In reality, most of the precipitation falling into the south-stretching watershed would be absorbed by the arid floodplain soils, with some of that water entering groundwater reservoirs, and some of that lost through evaporation up to about 3 mm per day, unless covered by soil and salt crusts [35] [36]. However, seepage and run-off from adjacent upslope areas would easily fill water-receiving depressions within floodplains, as appears to be the case for the coastal floodplain of Antofagasta ( Figure 13).
In quantitative terms, the DEM-derived estimates for the event-specific stream discharge rates (y, in m 3 ·sec −1 ) across the Antofagasta watershed areas vary by minimum water-retaining depression volumes (x, in GL) as follows: 2 10 307.9 34.8log ; 0.979 For the northern portion, overall depression-based water retention would be small, and would therefore do not depend in x in a major way, i.e.: These equations were obtained through GIS and regression analysis based on systematically determining y by decreasing the water-retention volume threshold per depression (x) from 10 GL to 0.1 ML. The largest water retaining volume capacity amounted to 0.28 GL in the northern section at the watershed divide leading eastward towards Salar del Carmen, and 4.22 GL in the southeastern floodplain at the watershed divide along the Pan American Highway, with International Journal of Intelligence Science Oficina Rosario located due east.
Further Comments The SRTM-DEM derived flow networks not only conform to varying degrees with already mapped river and stream delineations (GIS-DIVA), but also extend these delineations with greater accuracy towards and upwards into the many valleys of the Andean mountain range towards the east. Typically, the channel-to-channel distance conformance between already mapped and the SRTM-DEM derived flow channels is <100 m, 8 times out of 10 (details not shown). There is also a close correspondence between image and DEM-derived floodplain extent with the threshold for upslope floodplain flow-initiation area set at 400 ha. Terrace heights above the flow-channels with the floodplains can be varied, as these increase with increasing flow accumulation, up to about DTW = 40 m. In addition, within the Coquimbo region, valley soils tend to support vegetation growth up to about DTW = 40 m within the Coquimbo region.
Through the overlays and processing of precipitation patterns, watershed basins, floodplains, depressions flow channels, roads and railways, it is now possible to estimate: 1) maximum amounts of water transmitted for the upslope watershed areas at any flow-channel point of interest; 2) storm-average maximum run-off rates at any flow-channel location per duration of storm event; 3) approximate flood extent within the lower lying floodplain portions, done by varying the DTW threshold away from the floodplain stream channels; 4) DEM-determined water retention capacities of, e.g., water reservoirs, open-pit mines, quarries, and tailing ponds. For the context of any particular rainfall event, the information so derived can be used to gauge existing and required infrastructure requirements to withstand actual and projected storm events. Of general importance in this regard is determining the relationship between maximum average potential run-off rates per storm event and the corresponding water retention threshold based on, e.g., depressions and other water-retaining features within watersheds. For this purpose, and to achieve greater accuracies, it is necessary to determine basin-specific relationships between peak and average flow rates per storm duration by way of hydrometric calibrations. Doing so will assist in determining storm-and basin-specific run-off coefficients as they would change by storm event and by antecedent soil, groundwater, and reservoir conditions. For example, the June 2017 run-off coefficient pertaining to the Quebrada La Negra watershed south of Antofagasta was likely near zero, since the water would have mostly been depression-retained through soil and groundwater retention across the far-reaching upslope floodplain complex. Similarly, the March 2015 and May 2017 storm events may or may not have contributed water to Salar del Carmen east of Antofagasta.
Much additional progress in terms of hydro-spatial analyses will likely accrue through processing higher resolution DEMs, either obtained through fusing al-Journal of Geographic Information System ready existing DEM layers (e.g. SRTM, ASTER, elevation contours), and/or using airborne or satellite LiDAR-based bare-earth DEMs [37]. This information would not only enhance the accuracy of the above hydrographic delineations and related flood-exposure assessments but would also assist in quantifying the existing capacity of roads, railways, storm-water systems and other linear corridor infrastructure to withstand major storms. In this regard, road and railway delineations as available per, e.g., Open Street shapefiles generally coincide well with globally available surface images. In contrast, open-source stream and open-water delineations vary in detail and quality, and therefore may or may not follow hydro-topographic expectations in comprehensive detail.

Concluding Remarks
It is suggested that the DEM-based framework for guiding incoming precipitation through flow channels, floodplains, and depressions could find many practical as well as socio-economic applications to facilitate the planning and management of storm events and water supplies across northern Chile [38]- [43]. The approach as described and illustrated provides a quantitatively informing means to discuss, plan and evaluate hydrological interests and concerns at any location as these change from watershed to watershed, by storm event, and by upslope water-retaining capacities. To this effect, the framework generated results are limited in accuracy by the 30 m resolution of the SRTM-DEM, which-by way of grid-point interpolation-leads to inadvertent blocking of many flow channels, with the example of the Rio Loa gorge east of Calama in Figure 9 as an extreme case. Hence, the framework can, by itself, only be used to approximate how much storm water may flow and get trapped in real depressions. The amount so trapped is more readily quantified in large and deep depressions than in poorly defined depressions across flat to gently rolling terrain, as demonstrated via Equations (1) and (2) and exemplified in Table 2. In part, some of this difficulty can be addressed through manual if not automated breaching of artificial depressions. For the most part, however, greater accuracy can be achieved using DEMs with point resolution at 1 rather than 30 m. Doing so allows for systematic depression-to-depression water retention evaluations which could then be further evaluated by way of hydrometric and geophysical analyses, calibrations and validations.