3.1.1 Topography and catchment boundaries

The first step in the development of CAMELS-CL was the delimitation of catchment boundaries (Fig. 2). An official database for the Chilean hydrographic network was developed by the Instituto Geográfico Militar in 1984 (IGM, 1984) and updated by the DGA in 2014 (DGA and CIREN, 2014). This network delineation followed the Strahler hierarchy (Strahler, 1957), using the 30 m ASTER GDEM elevation data (Tachikawa et al., 2011). The DGA network includes 102 catchments, 491 sub-catchments and 1481 sub-sub-catchments, and has been largely used by government agencies, the private sector and the general public. However, a key limitation of this hydrographic network is that – given the methodology used for its implementation – the existing streamflow gauges do not correspond to catchment outlets. Furthermore, DGA catchment boundaries are truncated at the administrative national border, even though some catchments contribute runoff from the Bolivian and Argentinian territories. Since any hydrologic application within a controlled basin requires the total contributing area associated with streamflow measurements, and there is no official catchment boundary database, different studies have applied their own basin delineations, making it difficult to compare results.

To overcome this limitation, we created our own catchment boundary database for CAMELS-CL, defining the basin outlets at the location of 516 selected streamflow gauges (Sect. 3.1.5) and following only topographic-driven limits (not the administrative national border). A key challenge for this task is the mismatch between some station geographic coordinates reported by the DGA and the actual river network location – detected through the inspection of Google Earth imagery (Google, 2016). For some of those cases, expert advice was obtained from DGA technicians regarding gauge locations, while, for others, ancillary information (e.g. gauge name, road maps, Google Earth imagery) was used to determine the most likely location.

Basin delineation was performed in Quantum GIS (QGIS Development Team, 2015) by using watershed delineation packages from the Geographic Resources Analysis Support System (GRASS) (Neteler et al., 2012) and 30 m ASTER GDEM as input elevation data. Given the location of streamflow gauges, several catchments collected in this dataset are nested. We report this by using a logical hierarchy matrix indicating which basins are contained within another catchment of the dataset. The hierarchy matrix can be used to filter independent catchments, which is required for some applications such as hydrological modelling of large basins, catchment classification and parameter regionalisation.

The main topographic properties, including area, median, mean, minimum and maximum elevation, and mean slope, were computed for each catchment from ASTER GDEM 30 m raster data, clipped by the catchment boundary polygons and processed with the R raster package (Hijmans, 2016). An important limitation of this dataset is that its spatial resolution is relatively coarse, which can lead to errors when delineating catchments over very flat regions (such as the Far North; see Sect. 3.2.1).

3.1.2 Geology

Catchment-scale geological characteristics were retrieved from the Global Lithological Map database (GLiM) produced by Hartmann and Moosdorf (2012). GLiM is a compilation of national datasets into a unified global map. In the case of Chile, GLiM uses the map produced by the Servicio Nacional de Geología y Minería (Sernageomin, 2004), which has a resolution of 1:1 000 000 and is the most complete and commonly used map for the country. For each catchment, we reported the most and second-most frequent geological classes, as well as the fraction of the catchment they cover. We also extracted the fraction of the catchment described as “carbonate sedimentary rocks”, as it is a useful indicator of the presence of karstic systems.

3.1.3 Land cover

We used the 30 m resolution land cover map provided by Zhao et al. (2016), which integrates multi-seasonal Landsat 8 imagery acquired during 2013 and 2014, complemented with Moderate Resolution Imaging Spectroradiometer (MODIS) Enhanced Vegetation Index data, and high-resolution imagery on Google Earth. The classification scheme adopted by Zhao et al. (2016) was designed with Chilean geographers and biodiversity researchers, based on the FROM-GLC project (Gong et al., 2013), which is similar to the Land Cover Classification System (Di Gregorio and Jansen, 2005). This classification scheme is compatible with other land cover classification systems such as FAO or IGBP, with minor ancillary data. It consists of 10 main (level 1) classes (Fig. 1): croplands; forests, grasslands; shrublands; wetlands; water bodies; impervious surfaces; barren lands; and snow and ice. Some classes were refined in level 2 (e.g. forests were separated into native forest and exotic forest plantation) and level 3 subclasses (a total of 30 and 35 subclasses, respectively). For CAMELS-CL, we used the R raster package (Hijmans, 2016) to clip the land cover map within each catchment boundary polygon, and compute the fractional area associated with each class or subclass (as described in Table 3).

Figure 3Number of stations (left y axis) with at least f_d % of days with daily streamflow records, for different periods. The right y axis shows the percentage of catchments (out of 516) that meets the criterion. The period used in the CAMELS dataset (Addor et al., 2017) was included as a reference.

Download

3.1.4 Glaciers

Glaciers in Chile can be found at several locations, varying from small ice bodies at high-elevation sites in the Atacama region to alpine glaciers in the Central Zone and the large Patagonian ice fields in the Austral and Southern Patagonia regions (Pellicciotti et al., 2014). Even though the land cover map from Zhao et al. (2016) identifies areas of snow and ice, we included a global glacier inventory for calculating the degree of glacierisation of the selected catchments. Glacier inventories have the advantages of using geomorphologic glacier-delineation techniques and the recognition of debris-covered areas that cannot be identified by land cover classification schemes. In this study, we used the latest version of the Randolph Glacier Inventory (RGI 6.0; RGI Consortium, 2017). RGI 6.0 is a globally complete inventory of glacier outlines, and it is widely used in regional and global studies on land surface fluxes, climatology and meteorology (e.g. Huss and Hock, 2015; Marzeion et al., 2012; Mernild et al., 2017). We preferred to use RGI 6.0 rather than the Chilean glacier inventory from the DGA (DGA, 2014) because there are portions of some catchments lying in Argentinean territory (Fig. 2). RGI 6.0 was clipped within each catchment and two attributes were computed: the total glacierised area (km²) and the percentage of glacierised area in the catchment (%).

Figure 4Mean annual hydro-meteorological variables, calculated for the complete recording period of each variable. (b) and (d) were generated with precip_cr2met and pet_har products, respectively. The SWE product in (e) covers only the Near North and Central zones. The histograms indicate the number of catchments (out of 516) in each bin. The points represent the locations of catchment outlets.

Download

3.1.5 Streamflow

We compiled daily streamflow records for gauges maintained by the DGA, available from the CR2 Climate Explorer (http://explorador.cr2.cl/, last access: March 2018). From the 809 gauges included there, we selected those currently operational (independently of their data period) or suspended after 31 December 1980 with a record period longer than 10 years. We also discarded gauges located in artificial channels, ending up with 516 selected gauges. The record lengths from the selected gauges range from 192 to 366 667 days, with a mean (median) of 10 979 (9909) days. Figure 3 illustrates the availability of daily streamflow records for different time periods (represented by different colours), which is mainly due to the different record periods of the selected gauges. Further, sparse missing daily records may be found in streamflow time series due to specific extreme events, where the station might have not worked properly or might have been broken. Note that hydrological year is considered from 1 April to 31 March. As expected, the number of stations decreases with longer data availability. For example, if only stations with at most 5 % of missing data were selected, this would lead to a subset of 90 to 115 stations (which corresponds to 18 % and 22 % of the total number of catchments within the database, respectively), depending on the time period selected. When considering all stations with at most 30 % of missing data, then 249 to 258 stations (48 % and 50 % of catchments, respectively) would meet this criterion, depending on the period (Fig. 3). Figure 4a presents the mean annual discharge for each station (computed for the entire record period).

3.1.6 Precipitation

In most cases, precipitation is the main driver of hydrological systems. However, the geographical distribution of this variable is highly uncertain, even in densely monitored regions (Tian and Peters-Lidard, 2010; Woldemeskel et al., 2013). This limitation is aggravated in regions with difficult accessibility, where only a sparse network of meteorological stations is available. In order to account for robust precipitation estimates and to characterise the uncertainty of this variable, we processed catchment-scale precipitation from four different products, whose main characteristics are summarised in Table 1. Daily precipitation fields provided by each product were clipped and averaged within the catchment boundaries, resulting in four daily time series for each catchment, named precip_cr2met, precip_chirps, precip_mswep and precip_tmpa.

Table 1Precipitation products.

Download Print Version | Download XLSX

The precip_cr2met time series was derived from CR2MET, a spatially distributed daily precipitation product developed for Chile, which is currently used by DGA to update the national water balance (DGA, 2017). The CR2MET product is partly based on a statistical downscaling of ERA-Interim reanalysis data (Balsamo et al., 2015). The method builds on multiple linear regression models used to transfer precipitation, moisture fluxes and other variables from ERA-Interim onto 0.05^∘ resolution precipitation estimates. The statistical models, which also consider a number of topographic parameters, were calibrated using a large network of quality-controlled rain-gauge records. Depending on the distance of a given grid cell to neighbouring stations, the final product was obtained from merging downscaled precipitation and spatially interpolated in situ observations. Further information about formulation, quality control and product assessments can be found in DGA (2017).

The three satellite-based precipitation products used in CAMELS-CL were selected following the inter-comparison reported by Zambrano-Bigiarini et al. (2017) for the entire Chilean territory. The precip_chirps time series was computed from the Climate Hazards Group InfraRed Precipitation with Station data version 2 (CHIRPS, Funk et al., 2015), a long-term (1981 to near present), quasi global (50^∘ N to 50^∘ S) daily dataset available at a spatial resolution of 0.05^∘, designed to monitor agricultural drought and global environmental changes over land. CHIRPS uses the Tropical Rainfall Measuring Mission Multi-Satellite Precipitation Analysis version 7 (TRMM 3B42v7) in order to calibrate global cold cloud duration rainfall estimates (Funk et al., 2015). CHIRPS also incorporates surface rain-gauge data in order to reduce estimation biases, based on public and private monthly data. Originally, this dataset spanned from 50^∘ N to 50^∘ S, but since November 2012 data have not been produced south of 46^∘ S. More information can be found in Funk et al. (2015).

Variable precip_mswep was computed from Multi-Source Weighted-Ensemble Precipitation (MSWEP, Beck et al., 2017) data version 1.1, a fully global precipitation dataset released in June 2016, with 3-hourly temporal and 0.25^∘ spatial resolutions, specifically produced for hydrological modelling applications. MSWEP was designed to improve the performance of satellite products in representing precipitation over mountainous, tropical, and snowmelt-driven regions. The algorithm used in MSWEP merges observed rain-gauge data, satellite observations and reanalysis data to provide reliable precipitation estimates over the entire globe. In this paper, we used daily data from MSWEP version 1.1, but newer versions (already available) will be included in CAMELS-CL after validation with ground measurements in Chile.

Finally, precip_tmpa was computed from the Tropical Rainfall Measuring Mission (Huffman et al., 2007) Multi-satellite Precipitation Analysis (TMPA) dataset, which provides quasi global (50^∘ N to 50^∘ S) precipitation estimates at 0.25^∘ spatial resolution. TMPA integrates infrared and passive microwave data from a wide variety of satellite-borne precipitation-related sensors. In this study, we used TRMM research product 3B42v7, which makes use of Global Precipitation Climatology Project (GPCP; Adler et al., 2003) and Climate Assessment and Monitoring System (CAMS, Ropelewski et al., 1984) data to rescale its estimations on a monthly basis.

3.1.7 Temperature

Daily time series of minimum, maximum and mean temperatures for each catchment were also derived from the CR2MET dataset (DGA, 2017). Daily minimum and maximum temperatures in CR2MET (CR2MET/T_max and CR2MET/T_min, respectively) were mapped for the period 1979–2016 using a slightly different approach than the one used for precipitation (Sect. 3.1.6). In this case, the method used land surface temperature (LST) estimates from MODIS satellite retrievals, in addition to near-surface temperature provided by ERA-Interim. Multivariate regression models for both CR2MET/T_max and CR2MET/T_min were developed using LST as part of the explanatory variables and local temperature records in Chile as target data. Given the data gaps and relatively short period available for LST, the final product was derived for the whole period (1979–2016) by fitting the ERA-Interim data to the preliminary (incomplete) MODIS-based product. To get mean daily temperatures (CR2MET/T_mean), the long-term CR2MET/T_max and CR2MET/T_min were used to adjust the ERA-Interim 3-hourly near-surface temperature. The adjusted 3-hourly data were then averaged to derive CR2MET/T_mean. Gridded daily mean, minimum and maximum temperatures from CR2MET (0.05^∘ lat–lon resolution) were clipped to obtain basin-averaged daily time series for CAMELS-CL, named T_mean, T_min and T_max, respectively.

3.1.8 Potential evapotranspiration

We processed catchment-scale PET from two different sources. The first PET product uses the formulae proposed by Hargreaves and Samani (1985), which is solely based on surface temperature data (see Hargreaves and Allen, 2003, for further details). We used CR2MET/T_max and CR2MET/T_min (described in Sect. 3.1.7) to generate a gridded PET estimate (PET_har).

The second PET data included in CAMELS-CL (PET_mod) are those provided by the MODIS PET product (MOD16 collection 5; Mu et al., 2005), which is processed from different sources of information, including leaf area index and fractional photosynthetically active radiation, FPAR/LAI (MOD15A2; Myneni et al., 2002), land cover type 2 (MOD12Q1; Friedl et al., 2002), albedo (MCD43B2 and MCD43B3; Jin et al., 2003; Lucht et al., 2000), and daily meteorological reanalysis data from NASA's MERRA GMAO (GEOS-5). MOD16 is calculated following the Penman–Monteith approach (Howell and Evett, 2001), and the final product is available at an 8-day temporal resolution for the period 2000–2014, on a 1×1 km² grid. As for other gridded variables, the PET_har and PET_mod products were clipped and averaged within basin boundaries to generate daily (called pet_har) and 8-day (called pet_mod) catchment-scale time series, respectively.

3.1.9 Snow water equivalent

We processed daily snow water equivalent (SWE) data using the 180 m resolution SWE product generated by Cortés and Margulis (2017), which covers the Near North and Central zones (25–37^∘ S). Cortés and Margulis (2017) obtained SWE ensemble estimates from forward modelling “prior” values, which were conditioned through the assimilation of historical fractional snow-covered area (fSCA) data from Landsat TM, ETM+ and OLI sensors. The “posterior” SWE and fSCA estimates were probabilistically conditioned on the observed depletion record from Landsat, the uncertainty in fSCA observations, and the forward model state uncertainty. The fSCA retrieval was obtained with a spectral un-mixing algorithm (Cortés et al., 2014). The forward models for prior ensemble generation were the SSiB3 land surface model (Yang et al., 1997) and a snow depletion curve model (SDC; Liston, 2004). Detailed assessments of this reanalysis framework were performed for the Sierra Nevada using in situ sensor data (Margulis et al., 2016) and for the Andes (Cortés et al., 2016) using snow survey points, site years of peak annual snow pillow and snow course SWE observations. Validation results showed unbiased posterior SWE estimates with a correlation coefficient of 0.73, an RMSE of 0.29 m and a mean error of less than 0.01 m using snow pillow and snow course peak SWE. Results using snow survey data showed similar unbiased estimates, with a correlation coefficient of 0.50, an RMSE of 0.29 m and a mean error of less than 0.01 m. The daily SWE gridded product generated by Cortés and Margulis (2017) was clipped and averaged within the catchment boundaries to obtain daily time series for each catchment.

3.1.10 Water rights and reservoir information

A public reservoir dataset (http://www.ide.cl/descarga/capas/item/embalses-2016.html, last access: September 2017) was processed to identify the presence of dams within catchments. We also compiled and processed granted water rights available from the National Water Atlas (DGA, 2016a). This water allocation dataset includes information about the source (surface or groundwater), the type of right (i.e. consumptive or non-consumptive), use (i.e. industrial, irrigation, domestic and drinking water, hydroelectric power, pisciculture, mining, and classified as “other uses”), annual allocated flow (expressed in units of volume per time or as “shares”), and temporal allocation (i.e. permanent and continuous, permanent and discontinuous, permanent and alternated, eventual and continuous, eventual and discontinuous, or eventual and alternated). A detailed explanation of this water right classification can be found in Carey (2014). A key limitation of this dataset is the lack of information on the actual use of granted rights (Larraín, 2006). Additionally, some water right records have incomplete information (e.g. missing coordinates, water volume assigned and temporal allocation).

Figure 5 illustrates water allocation in central–southern Chile (30–43^∘ S), showing surface and groundwater rights (all types). It is clear that groundwater rights dominate in central Chile (31–36^∘ S), especially in low-elevation areas, compared to surface water rights. On the other hand, more surface water rights are granted in southern Chile, especially within high-elevation areas towards the Andes.

3.2.1 Location and topography

Figure 6 shows 6 (out of 14) location and topography attributes. Figure 6a presents the elevation of catchment outlets, illustrating two main elevation gradients: (i) a north to south (N–S) decrease, starting with high-elevation basins in the Far North macro-zone – which corresponds to the southern portion of the Altiplano Plateau (18–22^∘ S) (Allmendinger et al., 1997) – towards lower elevations in the southern macro-zones; and (ii) an east to west (E–W) gradient, dominated by high elevations in the Andes (located along the eastern border) decreasing towards sea level at the western border. This gauge elevation attribute can be used to classify catchments based on their location with respect to the coast or the Andes. We proposed the attribute location_type (see Table 3 and Fig. 6f) with three categories: coastal (or low elevation), foothills, and Altiplano catchments, defined by gauge elevations lower than 50 m a.s.l., between 1000 and 1200 m a.s.l., and above 3500 m a.s.l., respectively.

Figure 6Location and topography. For visualisation purposes, catchment areas (d) are shown up to their 90th percentile. The histograms indicate the number of catchments (out of 516) in each bin.

Download

Figure 6b reveals smoother N–S and E–W gradients of basin-averaged elevations, compared to gauge elevation gradients (Fig. 6a). This is because the mean elevation calculated for downstream catchments includes nested catchments (located at higher altitudes). From the complete set of catchments, 178 have a mean elevation greater than 2000 m a.s.l. The spatial distribution of mean catchment slopes follows different patterns depending on the macro-zone (Fig. 6c). The Far North – dominated by the flat Altiplano Plateau – exhibits relatively small variations in mean slopes, with relatively low values. From the Near North to Austral zones, the mean slope shows a spatial distribution similar to that from mean elevation, with an E–W gradient dominated by high slopes in the Andes and flatter areas towards the sea. In Southern Patagonia, such an E–W gradient is reversed given the relative position of the Andes.

The spatial distribution of basin areas shows a general increase from east to west (Fig. 6d), which is consistent with smaller headwater catchments at the Andes and larger downstream catchments towards the sea. Some exceptions to this E–W distribution pattern are catchments located near the eastern border, featuring either a N–S drainage direction or a portion of their total contributing area in Argentina (beyond the eastern national border). Additional exceptions to such E–W distribution are small inner sub-catchments near the western border or small headwater catchments originated at the Chilean Coastal Range, which runs from north to south along the Pacific coast and reaches up to 3000 m a.s.l. in the Antofagasta region (Figueroa and Moffat, 2000).

Given that all catchments were delineated using available streamflow gauge locations as outlets (Sect. 3.1.1), the contributing area (Fig. 6d) is not necessarily correlated with the number of nested basins within each catchment (Fig. 6e). For example, some small catchments might be highly instrumented (i.e. with many controlled nested basins, because of – for example – water allocation priorities or having a high population density like those in Region Metropolitana, which concentrates more than 40 % of the country's population), while large but poorly instrumented catchments might not have inner basins defined.

Figure 7Geology attributes. The histograms indicate the number of catchments (out of 516) in each bin.

Download

3.2.2 Geological attributes

Overall, the most common dominant geological classes within CAMELS-CL catchments are acid plutonic rocks (24 %), acid volcanic rocks (20 %) and pyroclastic (14 %). In the Far North zone, there is a strong presence of Pyroclastics, Siliclastic sedimentary rocks and Intermediate volcanic rocks (Fig. 7a and b), which can result in the connection of groundwater systems through fractured volcanic rocks (DGA, 1986). This means that there might be differences between surface catchment boundaries (based on topography) and the extension of groundwater systems, which should be considered when analysing basin-scale hydrological response. Figure 7a also indicates that strong geological differences may exist between neighbouring catchments. Furthermore, one can see generally high geological variability within the catchments. Indeed, the dominant geological class covers less than half of the contributing area in most catchments, as indicated by the histogram in Fig. 7c. The presence of carbonate sedimentary rocks is particularly low (Fig. 7e), with only 24 catchments having at least 10 % of this type of rock. This suggests low formation of karst, a subsurface characteristic featuring large fissures and voids, which results in fast infiltration rates and preferential permeability channels (La Moreaux et al., 1984).

3.2.3 Land cover attributes

As summarised in Table 2, five land cover attributes in A17 were not computed since the land cover map used here (from Zhao et al., 2016) does not provide information about leaf area index, green vegetation fraction, or depth. Instead, we included land cover attributes based on the catchment area encompassed by the main classes of the land cover dataset (Table 3). The first nine land cover attributes described in Table 3 were computed as the percentage of the catchment area covered by level 1 and 2 land cover classes defined by Zhao et al. (2016). We also computed a forest plantation index to quantify the ratio between forest exotic plantation (mainly Pinus radiata and Eucalyptus spp.) and native forest within a catchment, which is critical information for forest hydrology and ecosystem studies (e.g. Lara et al., 2009).

Considering that several catchments (almost 50; Fig. 1) extend beyond the Chilean territory, a key limitation of the land cover attributes derived from Zhao et al. (2016) is the lack of information outside the national boundary. To address this, we generated an attribute indicating the fractional catchment area contained within the land cover map, serving also as a quality flag for basin-averaged land cover characteristics.

Another limitation of the land cover map and derived attributes is that there is no characterisation of inter-annual variability (the map was constructed by using imagery from 2013 and 2014). This can be particularly important for land cover types that are sensitive to climatic variations, such as altiplanic wetlands, which largely influence the hydrology in the Far North. This limitation is also critical for classes featuring drastic changes within time, such as forest plantation and cropland.

Figure 8Land cover characteristics. Values below the colour bar lower limits are shown in blank (i.e. blank points in panel a represent no forest cover, and in e they represent no missing land cover data within those catchments). The histograms indicate the number of catchments (out of 516) in each bin.

Download

Figure 8 illustrates a subset of the land cover attributes listed in Table 3. Figure 8a shows the forested (native forests and forest plantation types) catchment area, which prevails in the Southern Zone, Austral Zone and Southern Patagonia. In forested catchments, exotic forest plantations dominate the coastal areas of the Central and Southern zones, with forest plantation indices up to 1 (Fig. 8b). Such a distribution illustrates the extensive land use change experienced in southern–central Chile over the last 50 years, where native forests have been progressively converted into agricultural lands and forest plantations (Armesto et al., 2010; Miranda et al., 2015). This conversion has had significant impacts on forest ecosystem services such as water provision (Jones et al., 2017; Lara et al., 2009).

Figure 8c and d show that the Far and Near North zones have more homogeneous land cover types, with shrublands and impervious lands occupying more than 60 % of the catchment areas. In southern areas, the coverage of the dominant classes decreases substantially, transitioning towards a mosaic of different land cover types. Missing land cover data are presented in Fig. 8e, which should be accounted for if the land cover attributes of the affected catchments (i.e. the ones with portions in Argentina, as shown in Fig. 2) are used for applications such as catchment classification or parameter regionalisation.

Because of the glaciological contributions to the water balance within the domain (Mernild et al., 2017; Le Quesne et al., 2009), we added two attributes (Table 3) based on information from the glacier inventory described in Sect. 3.1.4. We found that 255 catchments (48 % of the total) have some degree of glacierisation, reaching up to 62 % in the Geike River catchment in Southern Patagonia. The catchments with the largest degree of glacierisation (more than 15 %) are located in the Austral and Southern Patagonia regions, followed by the Olivares and Volcan river catchments (about 14 %) in the Central Zone. The glaciers included in CAMELS-CL span 7321 km², corresponding to almost a quarter of the glacierised area in the southern Andes (RGI Consortium, 2017).

3.2.4 Climatic indices

To allow direct comparisons between CAMELS (A17) and CAMELS-CL, climatic indices were computed for the same period as in A17, i.e. water years 1990 to 2009, corresponding to 1 April 1990 to 31 March 2010 for Chile. If these indices are required for different periods, the formulae provided in the references from Table 3 can be used with the raw hydro-meteorological time series (available from the CAMELS-CL website). The complete spatial and temporal coverages of meteorological variables allow the estimation of climatic indices for all 516 catchments – in contrast to hydrological signatures, computed for a subset of catchments (Sect. 3.2.5). Precipitation and PET-based attributes were calculated for all precipitation products (Sect. 3.1.6) using the daily PET product (pet_har, Sect. 3.1.8).

Table 3Summary of catchment attributes. Climate indices and hydrological signatures were computed for the period 1 January 1990–31 December 2010. Index i refers to the precipitation product (i=1, 2, 3, 4 for precip_cr2met, precip_tmpa, precip_chirps and precip_mswep, respectively).

Download XLSX

The climatic attributes presented in Figs. 4 and 9 reveal basic features of Chilean climatology, described in more detail by Miller (1976) and Garreaud et al. (2017), among others. Mean annual precipitation ranges from less than 10 mm yr⁻¹ in the Atacama Desert (northern Chile) to more than 3000 mm yr⁻¹ in western Patagonia (Fig. 4b). Such a marked precipitation gradient reflects the relative influence of the subtropical, semi-permanent south-eastern Pacific anticyclone, and the frequent incursion of frontal systems at higher latitudes. The frequency of high-precipitation events also increases southward, with a maximum in southern–central Chile (Fig. 9d). The Andean domain in the Far North (Chilean Altiplano), receiving about 300 mm yr⁻¹ above 4000 m a.s.l., is influenced by the monsoonal regime developing over the interior of the continent. In addition to the N–S gradient, precipitation increases strongly in the west–east direction due to the orographic enhancement of air masses over the windward slope of the Andes Cordillera (a factor of 2–3 between lowlands and windward slopes; Viale and Garreaud, 2014). PET has a more restricted range than precipitation (400–1400 mm yr⁻¹; Fig. 4d); therefore, the aridity index (PET/P, Fig. 9c) is higher in northern Chile (>1.0) compared to that of southern regions (<1.0). A positive precipitation seasonality (Fig. 9a) in northern Chile indicates precipitation peaks during summer (DJF), following the monsoonal precipitation regime governing in this region (Fig. 9f). In contrast, the negative seasonality values obtained for all macro-zones, except the Far North and Southern Patagonia, illustrate the increased storm frequency and high-precipitation events in most of the country during the winter (JJA) (Fig. 9a). Seasonality values close to zero indicate uniform precipitation throughout the year in Southern Patagonia (Fig. 9a and f). The zero-temperature isotherm during winter storms ranges between 1500 and 4000 m a.s.l., so most of the precipitation is liquid along the coast and interior valleys (Fig. 9b), while a larger fraction of solid precipitation is obtained in high-elevation basins.

Figure 9Climatic indices (calculated from the precip_cr2met product). The histograms indicate the number of catchments (out of 516) in each bin.

Download

3.2.5 Hydrological signatures

Hydrological signatures were computed for the period 1 April 1990 to 31 March 2010, as in Sect. 3.2.4. To exclude the effects of anthropic intervention on hydrologic response, we selected 94 catchments with valid daily streamflow records in at least 85 % of the period, based on the following criteria: interv_degree lower than 0.1 (i.e. less than 10 % of the annual streamflow allocated to surface rights), large_dam equal to zero (absence of large dams within the catchment), imp_frac lower than 5 % (negligible urban areas), copr_frac lower than 20 % (negligible irrigation effects) and fp_frac lower than 20 % (negligible forest plantation effects). Further, we excluded glacier-dominated catchments by selecting glacier_frac lower than 5 %. It should be noted that, despite calculating hydrological indices for a subset of catchments, raw daily time series for all 516 catchments are included in the CAMELS-CL database. These time series and the formulae provided in Table 3 may be used if the signatures are required for different time periods.

Figure 10 illustrates the spatial distribution of 12 (out of 14) hydrological signatures (Table 3), revealing the leading patterns of catchment responses. Both mean daily flow and runoff ratio increase from the Far North to the Southern Zone, showing strong correlations with mean annual precipitation (Fig. 4b) and the aridity index (Fig. 9c). Further, the runoff ratio shows a positive west to east gradient (i.e. increase towards the Andes), reaching values above 1 in mountain catchments in the Southern Zone. These non-behavioural catchments are further analysed in Sect. 4.1. Mean half-flow dates (Fig. 10c) present a similar west to east gradient, with higher values in steep (Fig. 6c) snow-dominated (Fig. 9b) basins in central Chile – where the most frequent season for low-precipitation days is DJF (Fig. 9i).

The mid-segment slope of the flow duration curve (FDC, Fig. 10d) – a signature that quantifies flashiness of runoff – shows that slow basin-averaged responses occur in the Far North and part of the Near North, in spatial correspondence to a high baseflow index (Fig. 10e) and low discharge precipitation elasticity (Fig. 10f). Such behaviour is expected in this region due to substantial subsurface and groundwater contributions to total runoff. Although flashiness of runoff and discharge elasticity to precipitation (baseflow index) are relatively higher (lower) and show some correlation towards the south, no clear spatial gradients are observed within the domain spanning from central Chile to Southern Patagonia. Figure 10f shows negative elasticity values (−0.13 and −0.03) in two catchments located in the Austral Zone and Southern Patagonia. Such values indicate a negative annual runoff anomaly in response to the positive annual precipitation anomaly, which is not expected in near-natural catchments. We attribute this behaviour to two main factors. First, there is a numerical problem in the formula used to calculate streamflow elasticity (Eq. 7 in Sankarasubramanian et al., 2001, adapted in A17, Table 3) when annual precipitation of a single year approaches the long-term mean, causing the elasticity to approach infinity (Sankarasubramanian et al., 2001). A second factor is the use of incomplete streamflow daily records. Since the elasticity is computed from concurrent daily streamflow and precipitation, its calculation in catchments with missing streamflow records can be problematic. This can be particularly important in snow-dominated catchments (delayed runoff response to precipitation) and in catchments with a weak precipitation seasonality (i.e. precipitation falling during the whole year, Fig. 9a close to zero), which is the case for catchments in the Austral Zone and Southern Patagonia.

The examination of signatures related to extreme (high or low) streamflow conditions exposes some interesting features. Although no clear spatial relationship is observed between high-flow signatures (Fig. 10g–i), similar spatial distributions of low-flow frequencies (Fig. 10j) and mean low-flow durations (Fig. 10k) are obtained across the country. Q₉₅ (Fig. 10i) and Q₅ (Fig. 10l) provide generally similar patterns to those of mean daily discharge (Fig. 10a), with positive increases from the Far North to Southern zones, and a positive west to east gradient. The comparison between the signatures displayed in Fig. 10g–l and climatic indices in Fig. 9d–i highlights the complex relationship between climate and hydrologic catchment behaviour. For example, the spatial structure in the frequency of low-/high-precipitation days is not reflected in the spatial distribution of high-/low-flow frequencies. A similar disjunction is observed between the duration of low-precipitation (Fig. 9h) and low-flow (Fig. 10k) events, whereas those catchments with low-duration of high-precipitation events also provide low durations in high-flow events.

Sharp variations in hydrological signatures (Fig. 10) – in contrast to generally smooth patterns in climate indices (Fig. 9) – are the result of complex, non-linear processes across a range of spatiotemporal scales, enhanced by heterogeneities in topography, soils, vegetation, geology and other landscape properties. Careful attention should be paid to such interactions and to the uncertainties involved in the calculation of hydrological signatures, in particular when attempting to extrapolate hydrological behaviour from gauged to ungauged basins based on climatic similarities alone (Westerberg et al., 2016; Westerberg and McMillan, 2015).

3.2.6 Human intervention

Figure 11 summarises water rights records used to characterise human intervention degree within the catchments. We can see that the number of surface rights (Fig. 11a) increases from north to south, while the number of groundwater rights (Fig. 11d) increases from east to west. Although these values do not provide information about allocated volumes, they show how many water rights holders interact to coordinate water use within a particular catchment. The CAMELS-CL database provides information about each water right within a catchment (not only the attributes with synthesised information), in case more detailed analyses are required.

In terms of allocated surface and groundwater flows (Fig. 11b and e, respectively), we only considered consumptive permanent water rights. Further, we considered only water rights recorded as volume per time, since water rights expressed as “shares” (6 % of the national water rights database) were not provided with their corresponding conversion into volume units (DGA, 2016b). It should be noted that shares rights are the oldest (allocated prior to the 2005 water code reform), thus probably representing a majority of the rights within the Central Zone (the region that concentrates the oldest rights).

Figure 10Hydrologic signatures for 94 near-natural catchments. The histograms indicate the number of catchments (out of 94) in each bin.

Download

The above limitations may lead to an underestimation of the allocated flow for (at least) the following reasons: (i) non-consumptive rights may have their restitution points outside the catchment boundaries (however, they were not considered for the allocated flow calculation); (ii) share rights are disregarded; (iii) there is missing information, and therefore some rights may be omitted (Sect. 3.1.10). On the other hand, allocation estimates may differ considerably from the actual extraction within a catchment. Possible reasons for this are the sub/over use of a granted allocated flow and unauthorised extractions of surface and groundwater.

Figure 11Water rights attributes. For visualisation purposes, the attributes in (a, b, d, e) are shown up to their 90th percentile. Attributes below the lower colour bar value are blank, and above the upper colour bar value they are blue. The histograms indicate the number of catchments (out of 516) in each bin.

Download

Despite the limitations of the water use dataset and the attributes presented in Fig. 11, water rights information is critical to quantify human intervention, and it has not been officially processed at the catchment scale in Chile. To quantify the intervention degree within a catchment, we calculated the interv_degree attribute (described in Table 3 and illustrated in Fig. 11c) as the ratio between the annual surface flow allocated within a catchment and the catchment mean annual runoff. This attribute indicates how much of the annual runoff generated – on average – within a catchment corresponds to the water volume allocated as consumptive surface rights. Further, we defined a binary attribute to characterise the presence of reservoirs within a catchment (large_dam in Table 3), using 0 if there are no dams and 1 if there is at least one dam. A limitation of these human intervention attributes (interv_degree and large_dam) that should be considered for hydrological applications is that they do not incorporate information about groundwater extractions.

To quantify the urbanised fraction of a catchment – another important factor modulating catchment response – we used the impervious fractional area attribute (imp_frac in Table 3), which usually contains urban areas. However, this land cover type is the worst classified class, since urban areas have mixed pixels of vegetation and paved surfaces (Zhao et al., 2016). The urban fraction of the catchments (assumed to be equal to imp_frac) varied between 0 % and 7 % for most catchments (only one catchment had imp_frac = 25 %).