BIOLOGICAL CONSERVATION 135 (2007) 500– 517

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/biocon

Freshwater conservation planning in data-poor areas: An example from a remote Amazonian basin (Madre de Dios River, and )

Michele Thiemea,*, Bernhard Lehnera,b, Robin Abella, Stephen K. Hamiltonc, Josef Kellndorferd,e, George Powella, Juan Carlos Riverosf aConservation Science Program, World Wildlife Fund, 1250 24th Street, NW, Washington, DC 20037, USA bDepartment of Geography, McGill University, 805 Sherbrooke Street West, Montreal, QC H3A 2K6, Canada cKellogg Biological Station and Department of Zoology, Michigan State University, 3700 East Gull Lake Drive, Hickory Corners, MI 49060, USA dRadiation Laboratory, Electrical Engineering and Computer Science Department, University of Michigan, 1301 Beal Ave, Ann Arbor, MI 48109, USA eWoods Hole Research Center, 149 Woods Hole Road, Falmouth, MA 02540, USA fWWF – Peru, Trinidad Mora´ n 853, Lince, Lima-14, Peru

ARTICLE INFO ABSTRACT

Article history: Using the 160,000-km2 drainage basin of the Madre de Dios and Orthon rivers in the south- Received 17 May 2006 west Amazon as a test case, we piloted an approach for large-scale conservation planning Received in revised form for freshwater systems characterized by a near-complete lack of biological and physical 24 October 2006 data. We used newly available spatial and remote sensing datasets, including spaceborne Accepted 26 October 2006 optical and radar observations, and new techniques of spatial data analysis to generate Available online 12 January 2007 subbasin (P100 km2), stream, and floodplain and wetland habitat classifications. We then generated a preliminary plan for a network of conservation areas to protect the most intact Keywords: examples of representative habitat types while maximizing longitudinal and lateral con- Representation nectivity. Proposed additions for freshwater conservation complement existing reserves Connectivity and build on earlier conservation planning efforts for terrestrial ecosystems. In the result- Floodplains ing integrated plan, at least 20% of the area of each major freshwater habitat type is repre- Habitat classification sented and two continuous corridors exist from the mouth of the Madre de Dios to its Bolivia headwaters in the . In total, we highlighted 84 currently unprotected subbasins to Peru fulfill our representation and connectivity goals. About two-thirds of these subbasins were considered relatively undisturbed and are identified as Level I (critical management zones) or Level II (indigenous territories), whereas one-third are potentially degraded and thus were designated as Level III (threat mitigation zones). This exercise provides an example of how newly available remote-sensing datasets and analytical tools may be used to advance freshwater conservation planning, particularly in data-poor regions. 2006 Elsevier Ltd. All rights reserved.

* Corresponding author: Tel.: +1 202 778 9760; fax: +1 202 293 9211. E-mail addresses: [email protected] (M. Thieme), [email protected] (B. Lehner), [email protected] (R. Abell), [email protected] (S.K. Hamilton), [email protected] (J. Kellndorfer), [email protected] (G. Powell), jc.riveros@wwfperu. org.pe (J.C. Riveros). 0006-3207/$ - see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.biocon.2006.10.054 BIOLOGICAL CONSERVATION 135 (2007) 500– 517 501

1. Introduction By nature of its vast size and remoteness, the suffers from additional information gaps. There are insuffi- The system presents both substantial oppor- cient hydrologic data to characterize the flow regime for large tunities and daunting challenges for conserving a globally portions of the basin, and biological data describing species outstanding share of the world’s freshwater biodiversity. distributions are lacking over much of the region. Even ‘focal The Amazon hosts the greatest richness of freshwater fish species’ like migratory catfish are so poorly known that design- species of any basin in the world, though precise numbers ing a conservation plan around their specific minimum habitat may be many decades off (Revenga et al., 1998; Goulding, requirements would be virtually impossible at this time, personal communication). Even less information is available though within a decade that information might be within reach for most other groups of aquatic taxa, but the sheer through new genetic analyses (Harvey and Carolsfeld, 2004). amount and great diversity of aquatic habitat suggest that These data gaps present serious obstacles to answering other biotic groups may be represented by high numbers the standard conservation planning questions, such as how as well. Among the Amazon’s most noteworthy but vulner- much area to protect, in what configuration, with what level able taxa are its long-distance migratory fishes; some spe- of habitat replication, and with what protection strategies. cies of large catfish may travel from the Amazon’s mouth This study builds on a growing body of work on freshwater to its headwaters over the course of their lifetimes (Goul- conservation planning (Roux et al., 2002; Smith et al., 2002; ding et al., 1996). Stein et al., 2002; Weitzell et al., 2003; Chadderton et al., Many opportunities for protecting freshwater species and 2004; Sowa et al., 2004) that provides guidance on these criti- their habitats exist, particularly in the western portion of cal questions, although all of these analyses are from regions the basin, where entire river systems are still relatively intact with significantly greater data availability than the Amazon. and where there are few large dams and other major struc- Addressing these questions within the Amazon will require tural changes to river channels (Nilsson et al., 2005). However, long-term scientific investigations, but the combination of development in the form of dams, commercial navigation existing opportunities and impending threats necessitates waterways, oil and gas extraction, and roads is on the hori- that the scientific community develop provisional hypotheses zon, and mining and logging concessions already cover large in the near-term that can be tested in the field over time. areas (Laurance et al., 2001; Maki et al., 2001; Fearnside, 2002; Within this context, we propose an approach for developing Steininger et al., 2002). Oil and gas exploration is active large-scale conservation plans for freshwater systems that throughout the Peruvian Amazon and extraction is expected are characterized by a near-complete lack of biological and to increase markedly over the next decade. Development physical data. We use the Madre de Dios Basin in the south- plans and concessions, while on the one hand ominous from west Amazon as a test case. This approach is exciting because a biodiversity conservation perspective, also present an it uses new datasets and analytical tools that we anticipate opportunity to infuse decisions proactively with sound envi- will have broad applicability and utility for freshwater conser- ronmental recommendations. vation in other data-poor regions around the world. Following There is growing evidence that the loss of functioning principles of conservation biology and freshwater ecology freshwater systems can have long-term ecological and eco- (Margules and Pressey, 2000; Higgins, 2003), the approach fo- nomic repercussions over large geographic areas (Holmlund cuses on protecting the most intact examples of representa- and Hammer, 1999; Gram et al., 2001). Costly fisheries de- tive habitat types and maximizing longitudinal and lateral clines following the construction of large dams around the connectivity through a network of conservation areas that world, such as the Tucuruı´ Dam on Brazil’s Tocantins River, complement existing reserves as well as terrestrial conserva- provide just one example (La Rovere and Mendes, 2000; World tion plans. Commission on Dams, 2000; Fearnside, 2001). South Amer- ica’s freshwaters are imperiled by a range of threats, of which 2. Methods dams are only the most obvious (Allan and Flecker, 1993; Prin- gle et al., 2000). In Amazonian headwater systems, threats 2.1. Region of analysis may actually derive as much or more from upland and ripar- ian land uses as from water engineering projects (Neill et al., The region of analysis is defined by the drainage basins of the 2001; Bernardes et al., 2004). Madre de Dios and Orthon rivers of Peru and Bolivia (hereafter The science to guide decision makers on how to protect referred to as the ‘‘Madre de Dios Basin’’; approximately Amazonian freshwater systems, either explicitly or within 160,000 km2; Fig. 1). This region is part of the Southwestern larger terrestrial initiatives, is inadequate. Some of the key Amazonian Moist Forests Global 200 ecoregion, which has scientific gaps are generic to freshwater systems around the been identified as outstanding for both terrestrial and fresh- world, whereas others are specific to the Amazon and other water biodiversity (Olson and Dinerstein, 1998). data-poor regions. Information gaps hindering nearly all Much of the runoff in the Madre de Dios system has its ori- freshwater conservation efforts include a poor understanding gin in the Andes. In particular, the Manu, Alto Madre de Dios, of species dispersal processes, metapopulation structures, Blanco, Azul, Colorado, Inambari, and Tambopata rivers drain and population viability; of species adaptations to and reli- from the Cordillera Oriental of the Andes (elevation approx. ance on natural regimes of flow and flooding; and how popu- 800–6300 m), which extends in a band covering the most lations, communities, and ecosystems are affected by southwestern edge of the region. Only the Orthon and Las Pie- interacting anthropogenic disturbances, including those orig- dras rivers lack drainage from the Andes. The rivers and inating on the terrestrial landscape (Abell, 2002). streams flowing through the Andes tend to be relatively clear 502 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

Fig. 1 – Major rivers within the drainage basin (gray) of the Madre de Dios and Orthon rivers. and highly responsive to rainfall. During the rainy season be- are common on poorly drained parts of upland alluvium tween October and April, they may rise and fall quickly and (Griscom and Ashton, 2003; Hamilton et al., in press). carry higher sediment loads than at other times. Andean trop- ical montane forests (also called yungas), which include cloud 2.2. Delineation of stream networks and subbasins forests, occur in a band from about 1000 to 3000 m. Above this elevation, sub-alpine forests grade into tropical alpine com- There is widespread agreement that drainage basins or water- munities (puna). Some streams running through cloud forest sheds are logical conservation planning and management tend to be more acidic (pH 4.2–5.2) than those running through units for freshwater biodiversity and water resource conser- other forest types (pH 6–7) (Goulding et al., 2003). vation, although some species will require larger spatial After descending from the Andes, the rivers start mean- scales than others and may use habitats within multiple ba- dering at about 350 m elevation. Extensive floodplains with sins (Wishart and Davies, 2003). Basins are nested hierarchi- oxbow lakes (cochas) occur along the courses of these lower cal units, and a larger basin can be subdivided into smaller elevation rivers. The rivers remain turbid throughout the year, basins that may serve as more tractable planning or manage- but carry their highest sediment loads during the rainy sea- ment units. These units need to be coarse enough in scale to son. Floodplain forests occur alongside the rivers; portions be useful for planning efforts across a large area, yet detailed of these forests are typically inundated several times between enough to make meaningful distinctions among freshwater December and April for brief periods (often <1–2 weeks). Large habitat types (FitzHugh, 2005; Higgins et al., 2005). We used areas of slightly more elevated floodplain also border the riv- subbasins, named such to distinguish them from the larger ers. These areas are seasonally flooded largely due to local Madre de Dios Basin, as our main unit of analysis, and our precipitation and poor drainage; they may infrequently or recommended areas for conservation intervention are built never receive floodwaters from the river. Along certain from these subbasins. reaches of the Madre de Dios and Tahuamanu rivers, palm Subbasins are particularly appropriate units of analysis in swamps or aguajales (dominated by Mauritia flexuosa L.) occur headwater regions, where a subbasin’s physical landscape in depressions on fluvial terraces and appear to receive little characteristics largely determine the characteristics of the or no water from the mainstem river. The waters of those streams draining it. On their passage downstream, however, palm swamps tend to be clear and acidic, in contrast to the small streams grow to larger rivers, which are increasingly higher turbidity and circumneutral pH of lowland rivers of influenced by upstream as well as local conditions. With the basin (Goulding et al., 2003). increasing distance from their source, the hydrologic charac- The lowland Madre de Dios basin has a humid tropical cli- teristics and ecological functionality of larger streams and mate (Osher and Buol, 1998) with rainfall varying from about rivers may therefore deviate more and more from the local 1200 to 3300 mm, generally increasing from east to west. conditions of a given subbasin through which they are pass- Rainfall is seasonal and lowest from June to September. Low- ing. To account for this distinction, we introduced a second land vegetation is predominantly evergreen or semi-ever- scheme of analysis units and explicitly distinguished polygo- green forest (Osher and Buol, 1998), and bamboo thickets nal subbasins from linear streams. Furthermore, floodplains BIOLOGICAL CONSERVATION 135 (2007) 500– 517 503

associated with larger lowland rivers comprise a distinct through a process of habitat classification that typically com- landscape unit that was considered separately. Thus our con- bines automated and manual processes. servation planning approach integrates features of the upland Numerous systems of aquatic classification have been drainage basins, the stream and river channels, and associ- developed around the world (Boon et al., 1998; Higgins et al., ated floodplains where they are large enough to warrant sep- 2005). Most systems are hierarchical and fall into two catego- arate treatment. ries: divisive or ‘top-down’ approaches or agglomerative or In principle, basin boundaries and stream lines can be im- ‘bottom-up’ approaches (Kingsford et al., 2005). ‘Top-down’ ported into a Geographic Information System (GIS) from exist- approaches start from large, ecologically heterogeneous ing maps. This method, however, produces results that vary areas, dividing these into lower more homogeneous levels. according to the scale and quality of the maps, and the prod- ‘Bottom-up’ approaches begin with the lowest levels of the uct is static. For the Amazon basin, the new digital elevation hierarchy and agglomerate them according to shared charac- data of the Shuttle Radar Topography Mission (SRTM; United teristics (Kingsford et al., 2005). Whereas it is generally agreed States Geological Survey, 2003) offers, for the first time, suffi- that understanding processes and patterns of freshwater sys- cient resolution (approx. 90 m horizontal, 1 m vertical) and tems at multiple scales is critical for conserving freshwater accuracy to derive reliable stream and basin delineations. biodiversity, there is still debate over which approach is best We obtained seamless SRTM data for the entire study area to classify these processes and patterns (Frissell et al., 1986; (downloaded from ftp://edcsgs9.cr.usgs.gov/pub/data/srtm/) Naiman et al., 1992; Fausch et al., 2002). For the data-poor and applied a series of ‘‘hydrologic enforcement’’ algorithms southwest Amazon the only option was a top-down ap- that conditioned the elevation data to enable improved calcu- proach, which we modeled on that developed by The Nature lations of flow directions and river topology. For these pro- Conservancy (Higgins et al., 2005). cesses we designed and programmed new, customized GIS tools, including algorithms for void filling, hydrologic filtering, 2.3.1. Subbasin and stream classification stream burning, stream carving, and scaling (Lehner et al., We developed the aquatic habitat classification scheme for 2006). We further developed tools and approaches that allow the Madre de Dios Basin through an iterative process. First, for an automated delineation of nested basins and stream we designed an initial classification framework by including networks from the conditioned SRTM data. These tools also all characteristics and combinations thereof that we consid- support the attribution of the resultant units with physical ered relevant to distinguish different habitat types, and for characteristics based on auxiliary data (e.g. slope, geology, which we had geospatial data. This initial scheme led to a and vegetation). Additionally, network topology was intro- large number (several hundred) of possible classes and sub- duced through a stream-order and coding system that per- classes, rendering it unmanageable for any practical applica- mits analyses of up- or down-stream connectivity, tion. To reduce the complexity of the classification scheme, distances, or relationships between any units within the we analyzed inter-correlations of the various parameters, study area. Visual comparisons of our results with existing identified the key variables, removed or grouped dependent detailed river maps of Peru (scale 1:100,000; Instituto Geogra´f- layers, and finally consolidated the initial framework into a ico Nacional, 1985–1989) and remote sensing imagery for se- feasible set of main classes. These steps and decisions were lected areas in Brazil (Side-Looking Airborne Radar; scale carried out manually and informed by consultation with sci- 1:1,000,000; Ministe´rio das Minas e Energia, 1982) showed entists familiar with the region. The specific classes and good overall agreement. We used a minimum size threshold thresholds were then reviewed at an expert workshop in of 100 km2 for our subbasin delineation in order to obtain a Lima, Peru in July 2004 and were further refined through fol- manageable number of units for planning at this scale. Due low-up consultations. We refrained from applying a multivar- to the configuration of the river network, the average subba- iate statistical analysis because we believe that a fully sin size within the study area was 250 km2, with the largest automated process – despite its objectivity – could neglect subbasin being 1350 km2. Streams were traced starting at all important influences like limited data quality or known outli- outlets of headwater subbasins and were subdivided into ers and thus lead to less desirable results than a user-super- individual reaches between all river confluences. In total we vised classification. distinguished 618 subbasins and 265 stream reaches. The classification framework for subbasins and streams focused on ecological functionality, which in turn is based 2.3. Classification of stream reaches, subbasins, and on hydrogeomorphic characteristics. The classification input floodplains categories were:

There is general agreement within the conservation commu- • System type (e.g., headwater vs. pass-through basin, small nity that protecting representative ecosystem types, or vs. large streams, floodplain type). ‘coarse-filter’ targets, should conserve common species and • Elevation (and derivatives, e.g., slope). communities, the ecological processes that support them, • Geology. and the environments in which they have evolved (Hunter, • Vegetation. 1991; Higgins et al., 2005). Distributional data for ‘fine-filter’ • Hydrology (e.g., surface runoff, river discharge). targets – typically rare, endemic, or endangered species – should ideally be used to complement coarse-filter Based on these categories we developed the initial hierar- information, but in regions like the Amazon existing species chical classification scheme: each type class was subdivided data are often insufficient. Coarse-filter targets are derived into different elevation zones, and each of these unique com- 504 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

binations was then further subdivided according to geology, Type of subbasin: vegetation, etc. In the following section, we describe the parameters and thresholds that we applied. Examples of attri- • Primary drainage basin Subbasin with largest stream butes that were tested initially but excluded from the final 6500 m3/s. classification are slope and river channel slope as derived • Pass-through basin Subbasin with a river >500 m3/s draining from elevation (both were highly correlated to elevation), through it. and river density (highly correlated to runoff). 2.3.1.2. Elevation. We distinguished three elevation zones, 2.3.1.1. Type. As described above, we were able to delineate representing the lowland plains, mid-elevation areas, and An- subbasins and stream reaches from the SRTM elevation data. dean mountain regions (Fig. 2a). The thresholds were again Floodplains and wetlands were subsequently delineated determined by experts, based on criteria to reflect both phys- based on SRTM plus additional remotely sensed imagery, as ical and biotic characteristics, including evidence that migra- described later. We distinguished three basic types of streams tory catfish spawn in the mid-elevation zone. Average and two basic types of subbasins. This classification is based elevation values were calculated from SRTM data for each on the size of streams and rivers in terms of their long-term subbasin, and maximum elevations were calculated for each average discharge, estimated via a hydrologic model as de- stream reach. The three elevation classes are: scribed later, because no measurements were available for this region. The thresholds were determined by consultation • Lowland 6300 m. with hydrologists and local experts. ‘‘Pass-through basins’’ • Mid-elevation >300 m and 6800 m. are adjacent to river channels and floodplains that receive • High-elevation >800 m. most of their water from upriver. Type of stream: 3 • Stream. Long-term average discharge 6500 m /s. 2.3.1.3. Geology. Geology can significantly influence the 3 • River. Long-term average discharge >500 m /s and chemical composition of surface runoff and groundwater. In 3 63000 m /s. general, the geological formations within the region of analy- 3 • Mainstem. Long-term average discharge >3000 m /s. sis are highly correlated with elevation. Yet there are some

Fig. 2 – Hydrogeomorphic characteristics of the Madre de Dios Basin: (a) elevation (SRTM data gridded at 90-m resolution); (b) stream reaches classified by % of upstream basin that occurs in the Andes (elevation > 800 m); (c) land cover (based on GLC2000: European Commission, 2003); and (d) model simulations of mean annual runoff (mm) per subbasin and mean annual discharge (m3/s) per stream reach. BIOLOGICAL CONSERVATION 135 (2007) 500– 517 505

special geological layers within the higher-elevation Andean 2.3.1.6. Final hydrogeomorphic/ecological classification scheme zone that appear to affect the hydrochemistry of their down- for subbasins and streams. Based on the initial hydrogeomor- stream drainages significantly (Barthem et al., 2003). Unfortu- phic/ecological classification scheme, we extracted a simpli- nately, we could not reliably identify these layers from the fied classification scheme for subbasins and streams. A few best region-wide geologic map (Geologic Data Systems, outliers were manually adjusted (e.g., single units differing 2003) and therefore introduced a simplified surrogate: we cal- from their homogenous neighborhood due only to a slight ex- culated the amount of high-elevation land area that drains to cess of a certain threshold). each stream reach (Fig. 2b) and classified the streams accord- Subbasin classes: ingly. Local experts recommended a threshold of 800 m as a suitable surrogate for Andean influence. The two surrogate • Pass-through basin (mainly along lowland rivers). classes for geology are: • Lowland wet tropical forest basin. • Lowland dry tropical forest basin. • With parts of the basin above 800 m elevation (more than • Mid-elevation tropical forest basin (mainly wet). 0%). • Mid-elevation bamboo forest basin (mainly wet). • Without parts of the basin above 800 m. • Mid-elevation montane forest basin (mainly wet). • High-elevation montane forest basin (mainly wet). • High-elevation grass- and shrubland basin (mainly dry). 2.3.1.4. Vegetation. We derived the dominant vegetation types per subbasin from the global GLC2000 land cover map (Euro- Stream classes: pean Commission, 2003; 1-km resolution; Fig. 2c). From the detailed map legend we distilled four basic classes: • Mainstem Madre de Dios River. • Lowland river with drainage from elevations >800 m (i.e. with • Bamboo forest. Andean influence). • Tropical forest. • Lowland river without Andean influence. • Montane forest. • Lowland stream with Andean influence. • Grass- and shrubland. • Lowland stream without Andean influence. • Mid-elevation stream with Andean influence. 2.3.1.5. Hydrology. Except for one known record of river • Mid-elevation stream without Andean influence. stage measurements collected at the Centro de Investigacio´ n • High-elevation stream (entirely Andean). y Capacitacio´ nRı´o de los Amigos (CICRA) field station above , Peru (Goulding et al., 2003; Hamilton 2.3.2. Floodplain classification et al., in press), there are no relevant hydrological data on Floodplains are a dominant and biologically critical feature runoff formation or river flows available for the region of within the region of analysis as well as within the wider Ama- analysis. Under this constraint, hydrological information zon basin (Puhakka et al., 1992; Lewis et al., 2000), and the com- had to be approximated by employing data on precipitation plexity of floodplain patterns cannot be captured sufficiently by and evapotranspiration to construct a hydrological model the subbasin approach. We use the term ‘‘floodplain’’ broadly to to simulate the spatio-temporal water balance. We utilized denote areas along rivers that are subject to at least occasional existing large-scale data from the Amazon-wide hydrologic inundation by overbank flow, as well as more distal areas that modeling system known as IBIS-HYDRA, produced for the may not be subject to riverine inundation (Hamilton et al., in Large-Scale Atmosphere–Biosphere Experiment in Amazoˆ nia press). The latter may have been created by fluvial action but project (Kucharik et al., 2000; Coe et al., 2002). The IBIS com- now lie on relatively uplifted terraces, or they may be areas sub- ponent of this model provides long-term monthly estimates ject to sheet flooding, often by water of local origin (i.e., rainfall of runoff generation at a grid resolution of 0.5 (approx. or local runoff). Inundation or at least soil saturation results be- 50 km). We disaggregated these values for the subbasin units cause they are level and poorly drained. They are nonetheless to derive estimates of average long-term runoff generation seasonally wet environments and in that respect would resem- for all subbasins (in mm/yr; Fig. 2d). We additionally accu- ble floodplains that are inundated by riverine overflow, mulated the runoff amounts along the stream network to although their biota could be quite distinct. calculate long-term means of river discharge. Without exist- We employed optical (Landsat ETM+) and microwave (JERS- ing discharge measurements the results of this modeling ap- 1 synthetic aperture radar) remote sensing data as well as the proach cannot be validated, but we believe that the SRTM topography to delineate floodplains and discern major estimates reflect at least the general hydrologic characteris- habitats within them (for details see Hamilton et al., in press). tics and trends of the study area. For the classification, we We analyzed the different spatial data sets simultaneously employed a simple division of ‘‘dry’’ and wet’’ areas, based with an object-oriented image analysis approach using the on consultation with hydrologists and experts familiar with software package ‘‘eCognition’’ (Professional Version 4, Defin- the region: iens Imaging, Munich). The object-oriented basis of this image analysis software proved to be superior in this case to tradi- • Dry. Average runoff generation within subbasin 61000 tional pixel-based image analysis for feature extraction and mm/yr. classification since contextual and topological rules can read- • Wet. Average runoff generation within subbasin >1000 ily be incorporated into a knowledge-based classification algo- mm/yr. rithm. The maps made by image analysis were field-checked 506 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

in a representative area along the Madre de Dios River, where riparian zone, and the natural ability of a river channel to mi- over 100 observations of habitat types were made and diverse grate laterally. The rich literature on riparian and floodplain waters on the floodplain were sampled for analysis of major buffers concludes that ideal widths are specific to individual solutes to reveal water sources (Hamilton et al., in press). Most systems (IUCN, 1992; Pringle and Benstead, 2001; Roux et al., of the field observations agreed well with the floodplain clas- 2002; Saunders et al., 2002). Bolivian and Peruvian forestry ses as delineated from remote sensing data, and the observed laws require 10–100 m and 50 m buffers, respectively, but differences were used to adjust the classification. Based on our these are not specific to the Amazon (Bolivia: Articulo 35 of field experience as well as published descriptions of the geo- the Reglamento General de la Ley Forestal; Peru: Articulo 26 morphology and vegetation of floodplains of the Peruvian of the Ley Forestal y de Fauna Silvestre). Along larger rivers Amazon (Kalliola et al., 1991; Puhakka et al., 1992; Pitman identified as important for meeting representation goals, we et al., 1999; Kvist and Nebel, 2001), we manually interpreted recommended that all adjacent subbasins be appropriately the detailed classification of floodplain environments to dis- managed to maintain buffer functions and connectivity and tinguish six key classes of major floodplains and wetlands also to conserve contiguous smaller-stream environments, for conservation planning purposes. The three main flood- which may support certain life history stages of the riverine plain classes (mainstem Madre de Dios, Manuripi River and fauna (e.g., spawning and nursery habitat for fishes and braided river floodplains) include a mosaic of other floodplain prawns). The lateral width of our subbasins varies, but the ef- environments such as palm swamps, meanderbelt forest, and fect was to create a buffer at least 1 km wide along these herbaceous vegetation that were consolidated for the pur- rivers. poses of conservation planning at this scale (see Supplemen- tary Map 1 for detailed floodplain classification). 2.5. Intactness assessment Floodplain and wetland classes: Although the region of analysis currently has a relatively low • Mainstem Madre de Dios River floodplain. human population density, there are several looming threats • Manuripi river floodplain. to the freshwater ecosystems of this basin. Among these, the • Braided river floodplains. most prominent is land conversion due to deforestation and • Palm swamps on fluvial terraces. agriculture (Goulding et al., 2003). Gold mining has had major • Palm swamps along river courses. impacts at a few sites, particularly at an area west of the • Pampas (savannas subject to seasonal flooding, generally Inambari River (Goulding et al., 2003). Smaller-scale gold due to poor drainage of rainfall rather than overbank flow mining also takes place in a more dispersed fashion along of river water; Hamilton et al., 2004). the rivers, especially on seasonally emergent bars. Mining- associated mercury contamination has caused concern, though the effects need further investigation (Guimara˜es 2.4. Connectivity et al., 1999; Lechler et al., 2000; Maurice-Bourgoin et al., 2000, 2003). Currently there are no large dams in the region Maintaining connectivity is a standard goal of conservation of analysis; however, river impoundment to allow navigation biology (Groves, 2003) and an essential and primary goal of through a reach of rapids on the downstream freshwater conservation planning (Abell et al., 2002). Our from the mouth of the Madre de Dios River is under consider- aim was to develop a plan that, if implemented, would protect ation. In addition to affecting the passage of aquatic animals longitudinal and lateral connectivity over entire river systems through that reach, upstream effects on the natural flow re- (Ward, 1989) and thereby permit the dispersal and migration gime might extend into the lower Madre de Dios and Orthon of aquatic species and the natural transfer of materials and rivers. energy (Federal Interagency Stream Restoration Working To identify areas that remain relatively intact, we overlaid Group, 1998). digital maps of population centers, roads, deforestation, and For longitudinal connectivity, we took an ‘all or none’ ap- mining with drainage and subbasin maps (WWF Peru digital proach: connectivity is either maintained through a complete data, J.C. Riveros, unpublished data). Population centers and lack of large dams (height P 15 m or volume > 3 million m3), roads provide good indicators of where habitat conversion or it is lost where such barriers exist anywhere along a river. and loss are likely concentrated, since human activities affect Large dams often affect the movements of fish and other the lands and waters surrounding these areas first (Sanderson organisms as well as disrupt key ecological processes such et al., 2002). Subbasins without population centers, roads, as flooding and sediment regimes (World Commission on land cover conversion, or mining activities were presumed Dams, 2000). We set a goal of maintaining longitudinal con- to contain relatively intact freshwater ecosystems. nectivity along the entire length of two major tributaries in Additionally, we assumed that areas under formal protec- the Madre de Dios Basin. As our region of analysis does not tion support relatively intact freshwaters. Few of these pro- extend to the mouth of the Amazon, we extended this goal tected areas have been designed to conserve freshwater only as far downstream as the mouth of the Madre de Dios; systems and species, but freshwater experts at a workshop however, if connectivity were disrupted downstream it would held in Lima, Peru in July, 2004 asserted that many of the affect migrations of aquatic species into the region (Barthem region’s large protected areas did confer protection to the and Goulding, 1997; Arajo-Lima and Ruffino, 2004). freshwaters running through them. We mapped protected Maintaining lateral connectivity requires protecting both areas designated as IUCN categories I–VI (nature reserves, the connection of the river to its tributaries, floodplain, and wilderness areas, national parks, natural monuments, BIOLOGICAL CONSERVATION 135 (2007) 500– 517 507

habitat/species management areas, protected landscapes, Oregon Department of Forestry et al., 2003). However, our goal managed resource protected areas), under the assumption of maximizing connectivity reduced the likelihood of recom- that these areas were most likely to be managed according mending a single, isolated subbasin for protection. to their conservation mandates. These areas cover We set a goal of capturing 20% of the area of each habitat 56,000 km2, or 35% of the region of analysis (Fig. 3). type in our portfolio of recommended conservation areas. Indigenous lands and territorial reserves are also expected Although this goal is widely-used in conservation (Murray to contribute to the integrity of the basin. While these areas et al., 1999; Noss et al., 2002; Sala et al., 2002; Airame et al., were not created with a biodiversity conservation goal, they 2003; Leslie et al., 2003; CSIR Environmentek, 2004), it is an represent formal barriers to the expansion of agriculture, important hypothesis to be tested over time with field data. Gi- cattle ranching and other forest conversion drivers. Recent ven that we have little understanding of the heterogeneity evidence provided by Nepstad et al. (2006) working in the among system types within the region, we chose not to set dif- Brazilian Amazon supports this argument. These areas ferential goals at this time, even though a higher degree of beta cover about 15,000 km2, or 9% of the region of analysis diversity is suspected in headwater streams than other aquatic (Fig. 3). habitats based on general patterns seen elsewhere (Jacobsen, 2003). Similarly, we applied our habitat representation goal 2.6. Representation analysis across the entire region of analysis, because we lack sufficient knowledge to subdivide the basin into biogeographically-in- Representation is defined loosely as conserving the full spec- formed stratification units (Ortega et al., in litt.). trum of biological and environmental variation. Setting repre- To conduct a gap analysis of freshwater habitat types with- sentation goals is intended to address the question of ‘how in the region, we overlaid existing and proposed protected much protection is enough?’ (Noss and Cooperrider, 1994; areas with the maps of classified floodplains, subbasins, and Groves, 2003). Without species or community data to inform streams, and then calculated the percent of each subbasin, our representation goals, we set goals based on those of sim- stream, and floodplain class that was captured within those ilar exercises described in the conservation planning litera- areas. Proposed protected areas for conserving terrestrial tar- ture, recognizing that most of the literature describes gets (Fig. 3) were those identified in WWF’s conservation plan conservation planning efforts in temperate environments. for the terrestrial biodiversity of the Southwestern Amazo- In terms of size, the subbasins that we delineated for our re- nian Moist Forests ecoregion (World Wildlife Fund, 2004). gion of analysis (mean area, 250 km2) are smaller than many For the river class calculation, all rivers were first buffered protected area size prescriptions for temperate freshwater out on both sides by 10 km, and full protection status was as- headwater systems (Thomas et al., 1993; Noss et al., 2002; signed only to those river reaches with more than 75% of their

Fig. 3 – Current protected areas (IUCN categories I–VI) and indigenous territories in the Madre de Dios Basin. Los Amigos is a private conservation concession under management of the Asociacio´ n para la Conservacio´n de la Cuenca Amazo´ nica (ACCA). 508 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

buffered area under protection. This calculation is an attempt 2.7. Selection of freshwater portfolio areas to adjust for the fact that river courses serving as the bound- aries of protected areas are not necessarily well-protected Our selection of freshwater portfolio areas was an iterative, themselves. largely manual process, grounded in the goals of achieving

Fig. 4 – Classification of subbasins and stream reaches within the Madre de Dios Basin. BIOLOGICAL CONSERVATION 135 (2007) 500– 517 509

representation, maximizing connectivity, and choosing the protection against instream barriers, or some degree of pro- most intact systems where possible. We did not use deci- tection of the terrestrial landscape beyond the riparian zone. sion-support software because no products were judged Priority areas that overlapped with indigenous territories able to address longitudinal connectivity adequately. We were classified as Level II, recognizing that achieving conser- assembled our portfolio of priorities using subbasins as vation objectives for these areas would depend on collabora- building blocks, to ensure that no boundaries cut through tion with indigenous groups. Where it was necessary to select subbasins. apparently disturbed areas to meet our representation and Consistent with the terrestrial selection process, we gave connectivity goals, we recommended these areas as Level first priority to areas that were adjacent to existing or pro- III. Level III areas experience high use and will require active posed protected areas, and that appeared relatively intact management and threat mitigation to contribute to conserva- based on interpretation of the threat data layers. From this tion of the basin’s freshwater systems. core set of areas we were then able to build out the portfolio to meet our goals. This process included selecting the head- 3. Results waters of existing and proposed protected areas to maintain longitudinal biophysical processes. We sought to minimize 3.1. Classification of stream reaches, subbasins, and the size of the selected portfolio by choosing areas that met floodplains multiple goals, where possible. To facilitate integration of our freshwater priorities with The subbasin and stream classification produced eight subba- the existing terrestrial conservation plan for the region, we sin classes and eight stream classes (Fig. 4). The initial, de- adopted a similar classification for priorities. Relatively intact tailed floodplain and wetland classification resulted in eight areas selected to meet our representation or connectivity classes (Supplementary Map 1). Six major floodplain and wet- goals were identified as Level I. Level I protected areas or crit- land types were subsequently derived from the detailed clas- ical management zones could be defined by a range of inter- sification for use in the representation analysis (Fig. 5). These ventions, including restrictions on use of the aquatic major types likely underestimate the diversity of floodplains system, the designation of floodplain or riparian reserves, and wetlands in the region, but remotely sensed information

Fig. 5 – Major floodplain and wetland types within the Madre de Dios Basin. This map depicts current river floodplains as well as relict floodplains and other poorly drained lands below 400 m elevation that are likely to support wetland ecosystems. These broad categories were distinguished based on a detailed classification (Supplementary Map 1) using remote sensing information. The ‘Other floodplain and wetland’ category was not included in the representation analysis because these areas lacked distinguishing features based on the remote sensing information that was available. 510 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

confirms that these types are distinct in their geomorphology, exception of some agricultural encroachment in the Tambo- hydrology, and vegetation. The remaining floodplain and wet- pata National Reserve and along the southwestern edge of land areas delineated in the detailed classification were not Manu National Park, it appears that the protected areas included in the conservation planning analysis because they have done relatively well at limiting intensive resource lacked distinguishing features based on available remote exploitation. sensing information. 3.3. Representation and connectivity gaps 3.2. Intactness assessment Despite the high level of protected area coverage within the Deforestation, mining, and human population centers are un- region of analysis, several freshwater habitat types are under- evenly distributed across the basin and are clustered around represented (Tables 1–3). One of the eight subbasin classes, major roads and rivers outside of protected areas and indige- five of the eight stream classes, and one of the six floodplain nous lands (Fig. 6). The road connecting Mazuco with Puerto classes are underrepresented under both the current and pro- Maldonado and then In˜ apari is a particularly disturbed corri- posed terrestrial protected area systems. An additional two dor that cuts across the mainstem Madre de Dios and its subbasin classes (lowland dry tropical forest and pass- floodplain. This area is due to come under far greater pres- through basins) would only be fully represented if the pro- sure, as the currently unpaved road is being upgraded to a posed terrestrial priority areas were implemented. Mid-eleva- major highway under the Regional Infrastructure Integration tion habitat types are well represented, but several of the low- Process of South America (IIRSA) initiative. Deforestation ap- and high-elevation classes are substantially lacking in pears most intense along this corridor, whereas mining and protection. agriculture occur both here and in more upstream areas. Longitudinal connectivity along the main rivers is fairly These observations concur with those from other parts of high within the region. However, there are a few major gaps. the Amazon Basin, where large-scale forest destruction has The section of the Madre de Dios mainstem that flows down- been linked with the construction of major new highways stream from its confluence with the Las Piedras River to (Laurance, 2000). Intensive gold mining continues along the Manuripi National Reserve currently lacks management or Madre de Dios upstream from Puerto Maldonado and along protection. Downstream from Manuripi to its confluence with the Inambari River about 145 km upstream from that same the Madeira River, the mainstem Madre de Dios also suffers point; during our field visits we observed disturbances of from a longitudinal gap in protection. A shorter gap exists the Madre de Dios river banks by placer mining. Agriculture along the Rio Tambopata from Puerto Maldonado to the also occurs in upstream portions of the Inambari Basin and northern edge of the Tambopata National Reserve. The largest between the Inambari and Madre de Dios rivers. With the longitudinal gap occurs along the mainstem Madre de Dios

Fig. 6 – Distribution of protected areas (IUCN categories I–VI), indigenous territories, and human activities in the Madre de Dios Basin. BIOLOGICAL CONSERVATION 135 (2007) 500– 517 511

Table 1 – Representation gaps among the subbasin habitat types of the drainage basin of the Madre de Dios River Subbasin habitat type Total area Existing protected Terrestrial nominated Freshwater nominated (km2) area (km2)/(%) area (km2)/(%) area (km2)/(%)

High-elevation grass- and shrubland 10,645 0 (0) 0 (0) 2164 (20) High-elevation montane forest 20,166 9522 (47) 0 (0) 2481 (12) Lowland dry tropical forest 44,829 7974 (18) 3634 (8) 3003 (7) Lowland wet tropical forest 11,997 6364 (53) 574 (5) 629 (3) Mid-elevation bamboo forest 11,291 9006 (80) 1814 (16) 0 (0) Mid-elevation montane forest 14,332 10,003 (70) 0 (0) 0 (0) Mid-elevation tropical forest 32,196 11,523 (36) 4552 (14) 2777 (9) Pass-through basin 13,824 1938 (14) 1123 (8) 5964 (43) Total 159,280 56,330 (35) 11,697 (7) 17,018 (11)

Percentages are calculated as a proportion of the total area of that habitat type. Those habitats that are underrepresented for existing protected areas are highlighted in italic. Those habitats that are underrepresented when both existing and nominated terrestrial protected areas are considered together are highlighted in bold.

Table 2 – Representation gaps among the stream habitat types of the drainage basin of the Madre de Dios River Stream habitat type Total area Existing Terrestrial Freshwater (km2 inside 10 km protected area nominated area nominated area buffer area) (km2)/(%) (km2)/(%) (km2)/(%)

High-elevation stream (Andean influence) 12,650 1259 (10) 0 (0) 1916 (15) Lowland river (Andean influence) 7475 651 (9) 0 (0) 1858 (25) Lowland river (no Andean influence) 6328 0 (0) 0 (0) 2394 (38) Lowland stream (Andean influence) 6578 3958 (60) 0 (0) 191 (3) Lowland stream (no Andean influence) 33,770 3037 (9) 0 (0) 4613 (14) Mainstem Madre de Dios 8625 69 (1) 522 (6) 3661 (42) Mid-elevation stream (Andean influence) 16,768 9960 (59) 0 (0) 160 (1) Mid-elevation stream (no Andean influence) 12,815 4086 (32) 2111 (16) 2977 (23) Total 105,009 23,020 (22) 2633 (3) 17,770 (17)

Percentages are calculated as a proportion of the total area of that habitat type. Those habitats that are underrepresented when both existing and nominated terrestrial protected areas are considered are highlighted in bold. For all calculations, rivers were first buffered out on both sides by 10 km, and full ‘‘protection’’ or ‘‘nomination’’ status is assigned to a river reach when its protected or nominated area exceeds 75% of the buffered area (see also text).

from Manu National Park downstream to the Manuripi Na- Madre Dios Rivers, respectively), the northern side of the Las tional Reserve. Piedras River (although the southern side is also only bordered Lateral connectivity gaps occur along many of the main riv- by a proposed protected area), and the northern edge of the ers that form the border of existing or proposed protected Tambopata National Reserve (Inambari and Tambopata rivers). areas. These include the northern and southern edges of There are also several ‘‘headwater gaps’’ – unprotected Manuripi National Reserve (bounded by the Manuripi and subbasins containing the most upstream headwaters of exist-

Table 3 – Representation gaps among the major floodplain types of the drainage basin of the Madre de Dios River

Floodplain habitat type Total Area Existing protected Terrestrial Freshwater (km2) area (km2)/(%) nominated area nominated area (km2)/(%) (km2)/(%)

Mainstem floodplain 7981 1577 (20) 424 (5) 3087 (39) Manuripi floodplain 799 438 (55) 0 (0) 345 (43) Braided floodplain 644 157 (24) 0 (0) 3 (0) Palm swamp on terraces 305 0 (0) 0 (0) 65 (21) Palm swamp-riverine 211 102 (48) 35 (17) 49 (23) Pampas 2810 537 (20) 0 (0) 0 (0) Total 12,750 2811 (22) 459 (4) 3549 (28)

Percentages are calculated as a proportion of the total area of that habitat type. Only the ‘‘palm swamp on terraces’’ habitat type (highlighted in bold) proved to be underrepresented when both existing and nominated terrestrial protected areas are considered together. Nevertheless, nominated areas would afford greater protection to the larger river floodplains. 512 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

Fig. 7 – Preliminary integrated conservation plan for the freshwater and terrestrial ecosystems of the Madre de Dios Basin. Level I contains relatively intact areas currently lacking protection, Level II areas overlap with indigenous lands, and Level III includes degraded areas or potential threat mitigation zones. Areas were first selected for their ability to meet representation and connectivity goals and then assigned to one of the three levels. Representation goals were met by Level I and Level III areas and connectivity goals were met by Levels II and Level III areas. ing protected areas. Four subbasins upstream of the Bahuaja cies within an area that is extremely data poor. Because the Sonene National Park were added to the portfolio to close this purpose of this exercise was primarily to test the creation gap. and analysis of the coarse filter, we did not apply a parallel and complementary fine-filter approach, intended to capture 3.4. Portfolio areas special elements-like rare and endemic species. In large part, these data are unavailable for the region, though some data In total, we highlighted 86 subbasins that require new or addi- exist for a few flagship species such as otters and side-necked tional protection to fulfill our representation and connectivity turtles. It may be a decade or more before the ranges of ende- goals. The areas we have recommended cover about mic or rare fish species are known throughout the region of 17,000 km2 in total. Of these areas, 44% are recommended analysis, and likely even longer for other freshwater taxa. as Level I, 29% are recommended as Level II, and 27% are rec- When better species-level data become available in the fu- ommended as Level III (Fig. 7). ture, they may be added to the analysis, thereby helping to The result is a preliminary integrated plan for the freshwa- identify habitats of particular importance as well as assist ter and terrestrial systems of the Madre de Dios Basin (Fig. 7). in setting more habitat-specific representation goals that cap- At least 20% of each freshwater habitat type is represented in ture differences in beta-diversity. In the meantime, a subse- this plan. Major corridors along the Las Piedras and Tambo- quent iteration of freshwater planning for this region might pata rivers also connect the most downstream reaches of incorporate measures of habitat complexity, derived from re- the Madre de Dios to its headwaters in the Alto Puru´ s Re- motely sensed imagery, to begin to identify areas of higher served Zone and to its headwaters in the Bahuaja Sonene Na- potential biodiversity. This information would not substitute tional Park, respectively. As new data become available within for fine-filter data, but could add a level of discrimination the region, this preliminary set of portfolio areas should be for areas within the same habitat class. revisited and revised if necessary. 4.2. Habitat classification 4. Discussion We present our habitat classifications as hypotheses that 4.1. Coarse-filter approach serve as a starting point for conservation planning in this re- mote region. Carefully designed stratified sampling in the Our coarse-filter habitat representation approach is intended field will be required to validate the classifications, and cor- to capture the majority of the more common freshwater spe- rections would have the potential to change the resulting BIOLOGICAL CONSERVATION 135 (2007) 500– 517 513

map of priorities considerably. We underscore, therefore, the were what we have termed ‘headwater gaps’ – that is, popu- highly provisional nature of our results. Nevertheless, this lated catchments that are upstream of existing and proposed study is of broader interest because it illustrates an approach protected areas. These areas present special challenges for to freshwater conservation planning that was previously not conservationists because of their potential impacts on down- possible. Our ability to generate a high-quality drainage stream quality. map from the new, high-resolution SRTM elevation data was The two main gaps in habitat class protection were for the critical starting point for a series of model applications high-elevation systems in the southern part of the Madre de that culminated in the classification of subbasins and river Dios Basin, and for nearly all of the lowland river classes, reaches. In a region with no river discharge data and no including the mainstem Madre de Dios River and its flood- pre-existing fine-scale subbasin maps, we believe that our re- plains. Gaps in lowland river protection are in fact a universal sults represent an important advance in freshwater conserva- problem for aquatic conservation, even though these systems tion planning that can be replicated elsewhere. Similarly, the provide critical habitats for both resident and migratory spe- mapping of floodplain and wetland habitats represents an cies (Peres and Terborgh, 1995) and their floodplains in partic- unprecedented combination of remote sensing observations ular comprise unique aquatic environments not found along and new image-analysis techniques. Hydrologic models and lower-order rivers and streams (Frissell, 1997). Much of the computer classification of remote sensing data should never disturbance in the region is concentrated in these lowland supersede field observations and data, but their outputs can gaps because the larger rivers of the Amazon have supported be used to design targeted sampling strategies, which are the transportation system of the region for hundreds of years. especially needed in vast and remote areas like the southwest Our freshwater plan, by recommending different types of pro- Amazon. tection, attempts to address these gaps in a practical fashion. The most obvious problem in terms of lateral connectivity 4.3. Applicability to other basins and ecoregions stems from the common practice of using rivers as the boundaries of protected areas. Protecting one bank is cer- The tools and approaches that have been developed to delin- tainly preferable to no protection at all, but implementing eate subbasins and stream networks, as well as to attribute floodplain or riparian management on the unprotected side these units with physical parameters, have been designed to would go far toward creating a buffer and providing habitat, work within a data-limited environment. The essential core both for aquatic and some terrestrial species (de Lima and layer of SRTM elevation data at 90-m resolution is available Gascon, 1999). On a related note, the practice of designating at near-global coverage and thus enables applicability in vir- rivers as protected area boundaries is questionable in a region tually all ecoregions or river basins globally. The tools we have where rivers may provide the only or best access points to developed will soon be available from the authors as exten- these areas; Peres and Terborgh (1995) argue for maximizing sions within a standard GIS software package (ArcView, Ver- the defensibility of Amazon nature reserves by using drainage sion 3.2, ESRI, Redlands, CA) and can thus be instrumental basins as the basis for delineating boundaries. in further analyses. Longitudinal connectivity gaps were surprisingly few, if we The classification framework and the required auxiliary assume that rivers flowing through or alongside protected data sets have been designed and selected in a generic man- areas are free from instream barriers. Those gaps that we ner and should therefore be transferable to other regions be- found, however, are critical ones. For example, the mainstem yond our region of analysis. The final classification scheme Madre de Dios downstream of Manu National Park is unpro- can be adapted to suit local requirements. The mapping of tected for about 150 km. Protecting the longitudinal connec- floodplains and wetlands based on globally available, space- tivity of priority rivers, both inside and outside protected borne remote sensing observations also can be extended to areas, would require a commitment by decision makers not other basins within the southwestern Amazon region and, to construct instream barriers on these systems. While the with appropriate calibration and validation, to other exten- risk of hydropower development may presently be low for riv- sive floodplain systems. ers in the region of analysis, a commitment by the govern- ments of Peru and Bolivia to protect migration corridors 4.4. Gaps in protection for freshwaters from the headwaters of the Las Piedras and Tambopata rivers down to the mouth of the Madre de Dios could serve as a Freshwater systems appear to be moderately well-protected strong argument to Brazil to reconsider plans for large-scale in the region of analysis, particularly when proposed terres- hydropower and navigation projects directly downstream trial priority areas are considered. The combined size of pro- (Fearnside, 2005). tected areas in the region and the fact that many were The final category of ‘headwater gaps’ highlighted in- originally designed to capture entire watersheds means that stances where protected areas failed to achieve their full po- rivers should receive some degree of protection no matter tential for conserving the freshwater systems running what the configuration of areas. However, despite the fact through them, because relatively small pieces of upstream that existing protected areas, as defined in this assessment, headwater catchments fell outside the areas. In some cases, cover 35% of the region of analysis, we found some notable remedying this problem could be accomplished by extending gaps in protection of the region’s freshwaters. These gaps the protected area boundaries. In other situations, the win- took three main forms. First, there were representation gaps dow of opportunity for protecting an entire intact drainage in protection of certain broad habitat classes. Second, there basin has passed, and attention must shift to management were lateral and longitudinal connectivity gaps. Last, there within a human dominated landscape. The importance of 514 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

implementing best management practices in headwaters, can nonetheless be used to catalyze conversations about how though, should not be underestimated, as these areas can to protect freshwater biodiversity in the Amazonian headwa- have disproportionately large influences on downstream ters and in other data-poor systems. The data sources used to water quality, nutrient regimes, and flow (Meyer and Wallace, complete our analysis are now available for most of the globe, 2001). which means that with a few relatively simple GIS and image- analysis tools, coarse filter conservation planning for fresh- 4.5. Selection of portfolio water biodiversity is virtually a universal possibility.

The process that we used for selecting priority areas allowed Acknowledgements us to meet representation and connectivity goals while also taking into account existing threats and terrestrial nominated We gratefully acknowledge the contributions of participants areas. In total about 11% of the region of analysis was nomi- at two workshops, whose results formed the foundation for nated for some level of protection, 57% for representation this work. These individuals are: Andrea Blanco, Ronaldo Bar- and 43% for connectivity. In the future, the development of them, Carlos Can˜ as, Fernando Carvajal, Jorge Celi, Bruce Fors- a decision-support software that explicitly addresses longitu- berg, Michael Goulding, Michael McClain, Bob Meade, Mariana dinal connectivity could allow for the generation of multiple Montoya, Hernan Ortega, Sue Palminteri, Fernando Regal, optimum sets that meet specific area requirements while Warren Rider, Tom Saunders, Carlos Scaramuzza, Mathias To- minimizing area. bler, Paul van Damme, and Lindsey Waggoner. This project has been supported with funds from the Gordon and Betty 5. Conclusions Moore Foundation and JohnsonDiversey Inc.

Our proposed plan is built on a combination of common Appendix A. Supplementary data sense, expert opinion, and sophisticated spatial data analysis and modeling. The implementation of these recommenda- Supplementary data associated with this article can be found, tions will require extensive stakeholder involvement and con- in the online version, at doi:10.1016/j.biocon.2006.10.054. siderations of social and economic tradeoffs. We recognize that creating additional protected areas in this already well- protected area would be challenging, and that much of the land outside protected areas and indigenous lands is already REFERENCES settled to various degrees. Consistent with the terrestrial plan for this region, we have designed a plan that recognizes that Abell, R., 2002. Conservation biology for the biodiversity protected areas are only one of several important and com- crisis: a freshwater follow-up. Conservation Biology 16, plementary strategies. In this region where people directly de- 1435–1437. pend on healthy freshwater systems and the services that Abell, R., Thieme, M., Dinerstein, E., Olson, D., 2002. A sourcebook they provide for their sustenance (Kvist and Nebel, 2001), for conducting biological assessments and developing biodiversity visions for ecoregion conservation. In: Freshwater there are strong arguments to be made for mitigating distur- Ecoregions, vol. II, World Wildlife Fund, Washington, DC. bances that impair freshwater systems and for working with Airame, S., Dugan, J.E., Lafferty, K.D., Leslie, H., McArdle, D.A., communities to develop multiple-use management zones. Warner, R.R., 2003. Applying ecological criteria to marine Relatively simple measures to safeguard migratory fish corri- reserve design: a case study from the California Channel dors and the floodplains that nourish countless species and Islands. Ecological Applications 13, S170–S184. valuable fisheries are also well-justified from an ecosystem Allan, J.D., Flecker, A.S., 1993. Biodiversity conservation in services perspective. running waters. BioScience 43, 32–42. Arajo-Lima, C.A.R.M., Ruffino, M.L., 2004. Migratory fishes of the Key to creating an effective conservation plan is being able Brazilian Amazon, In: Carolsfield, J., Harvey, B., Ross, C., Baer, to visualize water and land together. Terrestrial conservation A. (Eds.), Migratory fishes of South America: Biology, Fisheries planners have more often than not created blueprints without and Conservation Status, World Fisheries Trust/World Bank/ undertaking the simple act of overlaying rivers with their land IDRC, pp. 233–302. cover and land use maps. To be fair, in some cases only very Barthem, R., Goulding, M., 1997. The Catfish Connection: Ecology, poor river maps have been available; a new global drainage Migration, and Conservation of Amazon Predators. Columbia University Press, New York. map (Lehner et al., 2006) will help address that problem. Even Barthem, R., Goulding, M., Forsberg, B., Can˜ as, C., Ortega, H., 2003. where no freshwater classifications are available, applying Aquatic Ecology of the Rı´o Madre de Dios: Scientific Bases for simple rules of thumb, like encompassing entire drainage ba- Andes-Amazon Headwaters Conservation. Asociacio´ n para la sins and promoting lateral and longitudinal connectivity Conservacio´ n de la Cuenca Amazo´ nica/Amazon Conservation along rivers, can make any plan more ‘freshwater friendly.’ Association, Lima, Peru. Truly integrated terrestrial-freshwater conservation planning Bernardes, M.C., Martinelli, L.A., Krusche, A.V., Gudeman, J., would require simultaneous identification of priority systems Moreira, M., Victoria, R.L., Ometto, J.P.H.B., Ballester, M.V.R., Aufdenkampe, A.K., Richey, J.E., Hedges, J.I., 2004. Riverine in the two realms, and subsequent recommendations based organic matter composition as a function of land use changes, on the most efficient intersection of the two priority sets. In southwest Amazon. Ecological Applications 14, S263– this example of partial integration, we have been able to take S279. planning substantially further than was previously possible to Boon, P., Wilkinson, J., Martin, J., 1998. The application of SERCON produce a scientifically grounded map that, while provisional, (System for Evaluating Rivers for Conservation) to a selection BIOLOGICAL CONSERVATION 135 (2007) 500– 517 515

of rivers in Britain. Aquatic Conservation: Marine and environmental samples from two lakes in Amapa´, Brazilian Freshwater Ecosystems 8, 597–616. Amazon. Ambio 28, 296–301. Chadderton, W.L., Brown, D.J., Stephens, R.T., 2004. Identifying Hamilton, S., Sippel, S., Melack, J., 2004. Seasonal inundation Freshwater Ecosystems of National Importance for patterns in two large savanna floodplains of South America: Biodiversity: Criteria, Methods, and Candidate List of the Llanos de Moxos (Bolivia) and the Llanos del Orinoco Nationally Important Rivers. Department of Conservation, (Venezuela and Colombia). Hydrological Processes 18, 2103– Wellington, New Zealand. Available from: www.doc.govt.nz/ 2116. Publications/004~Science-and-Research/Miscellaneous/PDF/ Hamilton, S.K., Kellndorfer, J., Lehner, B., Tobler, M., in press. WONI.pdf. Remote sensing of floodplain geomorphology as a surrogate Coe, M.T., Costa, M.H., Botta, A., Birkett, C., 2002. Long-term for habitat diversity in a tropical river system (Madre de Dios, simulations of discharge and floods in the Amazon basin. Peru). Geomorphology. Journal of Geophysical Research 107, 8044. Harvey, B., Carolsfeld, J., 2004. Introduction: Fishes of the CSIR Environmentek, 2004. National Spatial Biodiversity floods. In: Carolsfeld, J., Harvey, B., Baer, A., Ross, C. (Eds.), Assessment: Rivers Component. Technical Report. CSIR report Migratory Fishes of South America: Biology, Social Importance, number ENV-S-I-2004-063. CSIR, Stellenbosch, South Africa. and Conservation Status. World Fisheries Trust, Canada, de Lima, M., Gascon, C., 1999. The conservation value of linear pp. 1–18. forest remnants in central Amazonia. Biological Conservation Higgins, J.V., 2003. Maintaining the ebbs and flows of the 91, 241–247. landscape: conservation planning for freshwater ecosystems. European Commission, 2003. Global Land Cover 2000 Database. In: Groves, C. (Ed.), Drafting a Conservation Blueprint: A European Commission, Joint Research Centre, Ispra, Italy. Practitioner’s Guide to Planning for Biodiversity. The Nature Available from: http://www.gvm.jrc.it/glc2000/. Conservancy and Island Press, Washington, DC, pp. 291–318. Fausch, K.D., Torgersen, C.E., Baxter, C.V., Li, H.W., 2002. Higgins, J.V., Bryer, M.T., Khoury, M.L., FitzHugh, T.W., 2005. A Landscapes to riverscapes: bridging the gap between research freshwater classification approach for biodiversity and conservation of stream fishes. BioScience 52, 483–498. conservation planning. Conservation Biology 19, 432–445. Fearnside, P.M., 2001. Environmental impacts of Brazil’s Tucuruı´ Holmlund, C.M., Hammer, M., 1999. Ecosystem services generated Dam: unlearned lessons for hydroelectric development in by fish populations. Ecological Economics 29, 253–268. Amazonia. Environmental Management 27, 377–396. Hunter Jr., M.L., 1991. Coping with ignorance: the coarse filter Fearnside, P.M., 2002. Avanc¸a Brasil: Environmental and social strategy for maintaining biodiversity. In: Kohm, K. (Ed.), consequences of Brazil’s planned infrastructure in Amazonia. Balancing on the Brink of Extinction: The Endangered Species Environmental Management 30, 735–747. Act and Lessons for the Future. Island Press, Washington, DC, Fearnside, P.M., 2005. Brazil’s Samuel Dam: lessons for pp. 266–281. hydroelectric development policy and the environment in Instituto Geogra´fico Nacional, 1985–1989. Cartas Nacionales Amazonia. Environmental Management 35, 1–19. (National Charts). Scale 1:100,000. Repu´ blica del Peru´ , Lima, Federal Interagency Stream Restoration Working Group, 1998. Peru´. Stream Corridor Restoration: Principles, Processes, and IUCN, 1992. Conserving biological diversity in managed tropical Practices. National Technical Information Service, Springfield, forests. In: Proceedings of a Workshop Held at the IUCN VA. General Assembly, Perth, Australia, 30 November to 1 FitzHugh, T.W., 2005. GIS tools for freshwater biodiversity December 1990, IUCN, Gland, Switzerland and Cambridge, UK. conservation planning. Transactions in GIS 9, 247–263. Jacobsen, D., 2003. Altitudinal changes in diversity of Frissell, C.A., 1997. Ecological principles. In: Williams, J.E., Wood, macroinvertebrates from small streams in the Ecuadorian C.A., Dombeck, M.P. (Eds.), Watershed Restoration: Principles Andes. Archiv fuer Hydrobiologie 158, 145–167. and Practices. American Fisheries Society, Bethesda, Kalliola, R., Puhakka, M.E., Salo, J., Tuomisto, H., Ruokolainen, K., Maryland, pp. 96–115. 1991. The dynamics, distribution and classification of swamp Frissell, C.A., Liss, W.J., Warren, C.E., Hurley, M.D., 1986. A vegetation in . Annales Botanici Fennici 28, hierarchical framework for stream habitat classification: 225–239. viewing streams in a watershed context. Environmental Kingsford, R.T., Dunn, H., Love, D., Nevill, J., Stein, J., Tait, J., 2005. Management 10, 199–214. Protecting Australia’s Rivers, Wetlands and Estuaries of High Geologic Data Systems, 2003. Computerized geologic mapping Conservation Value: A Blueprint. Land and Water Australia, program of South America. Available from: http:// Canberra, Australia. Available from: http://www.ids.org.au/ www.gdata.com. ~cnevill/RiversBlueprint.pdf. Goulding, M., Smith, N.J.H., Mahar, D.J., 1996. Floods of Fortune: Kucharik, C.J., Foley, J.A., Delire, C., Fisher, V.A., Coe, M.T., Gower, Ecology and Economy along the Amazon. Columbia University S.T., Lenters, J., Molling, C., Norman, J.M., Ramankutty, N., Press, New York, NY. 2000. Testing the performance of a dynamic global ecosystem Goulding, M., Can˜ as, C., Barthem, R., Forsberg, B., Ortega, H., 2003. model: water balance, carbon balance, and vegetation Amazon Headwaters: Rivers, Wildlife, and Conservation in structure. Global Biogeochemical Cycles 14, 795–825. Southeastern Peru. Amazon Conservation Association, Lima, Kvist, L.P., Nebel, G., 2001. A review of Peruvian flood plain forests: Peru. ecosystems, inhabitants and resource use. Forest Ecology and Gram, S., Kvist, L.P., Ca´seres, A., 2001. The economic importance Management 150, 3–26. of products extracted from Amazonian flood plain forests. La Rovere, E.L., Mendes, F.E., 2000. WCD Case Study: Tucuruı´ Ambio 30, 365–368. Hydropower Complex, Brazil. Prepared for the World Griscom, B.W., Ashton, M.S., 2003. Bamboo control of forest Commission on Dams Secretariat. Available from: http:// succession: Guadua sarcocarpa Pilger in southeastern Peru. www.dams.org/docs/kbase/studies/csbrmain.pdf. Forest Ecology and Management 175, 445–454. Laurance, W.F., 2000. Mega-development trends in the Amazon: Groves, C., 2003. Drafting a Conservation Blueprint: A implications for global change. Environmental Monitoring and Practitioner’s Guide to Planning for Biodiversity. Island Press, Assessment 61, 113–122. Washington, DC. Laurance, W.F., Cochrane, M.A., Bergen, S., Fearnside, P.M., Guimara˜es, J.R.D., Fostier, A.H., Forti, M.C., Melfi, A.J., Kehrig, H., Delamoˆ nica, P., Barber, C., D’Angelo, S., Fernandes, T., 2001. Mauro, J.B.N., Maim, O., Krug, F.J., 1999. Mercury in human and The future of the Brazilian Amazon. Science 291, 438–439. 516 BIOLOGICAL CONSERVATION 135 (2007) 500– 517

Lechler, P.J., Miller, J.R., Lacerda, L.D., Vinson, D., Bonzongo, J.C., Oregon Department of Forestry, Oregon Department of Fish and Lyons, W.B., Warwick, J.J., 2000. Elevated mercury Wildlife, Oregon Trout, 2003. Salmon Anchor Habitats Strategy concentrations in soils, sediments, water, and fish of the for Northwest Oregon State Forests. Available from: Madeira River basin, Brazilian Amazon: a function of natural www.odf.state.or.us/DIVISIONS/management/state_forests/ enrichments? The Science of the Total Environment 260, 87– sfplan/ip03_finals/13_SAH.pdf. 96. Osher, L.J., Buol, S.W., 1998. Relationship of soil properties to Lehner, B., Verdin, K., Jarvis, A., 2006. HydroSHEDS Technical parent material and landscape position in eastern Madre de Documentation, Version 1.0. World Wildlife Fund, Dios, Peru. Geoderma 83, 143–166. Washington, DC. Available from: www.worldwildlife.org/ Peres, C.A., Terborgh, J.W., 1995. Amazonian nature reserves: an hydrosheds. analysis of the defensibility status of existing conservation Leslie, H., Ruckelshaus, M., Ball, I.R., Andelman, S., Possingham, units and design criteria for the future. Conservation Biology H.P., 2003. Using siting algorithms in the design of marine 9, 34–46. reserve networks. Ecological Applications 13, S185–S198. Pitman, N.C.A., Terborgh, J.W., Silman, M.R., Nunez, P., 1999. Tree Lewis, W.M., Hamilton, S.K., Lasi, M.A., Rodriguez, M., Saunders species distributions in an upper Amazonian forest. Ecology Iii, J.F., 2000. Ecological determinism on the Orinoco floodplain. 80, 2651–2661. BioScience 50, 681–692. Pringle, C.M., Benstead, J.P., 2001. The effects of logging on tropical Maki, S., Kalliola, R., Vuorinen, K., 2001. Road construction in the river ecosystems. In: Fimbel, R.A., Grajal, A., Robinson, J.G. Peruvian Amazon: process, causes and consequences. (Eds.), The Cutting Edge: Conserving Wildlife in Logged Environmental Conservation 28, 199–214. Tropical Forest. Columbia University Press, New York, NY, pp. Margules, C.R., Pressey, R.L., 2000. Systematic conservation 305–325. planning. Nature (London) 405, 243–253. Pringle, C.M., Scatena, F.N., Paaby-Hansen, P., Nu´n˜ ez-Ferrera, M., Maurice-Bourgoin, L., Quiroga, I., Chincheros, J., Courau, P., 2000. Boon, P.J., 2000. River conservation in Latin America and the Mercury distribution in waters and fishes of the upper Madeira Caribbean. In: Davies, B.R., Petts, G.E. (Eds.), Global rivers and mercury exposure in riparian Amazonian Perspectives on River Conservation: Science, Policy and populations. The Science of the Total Environment 260, 73–86. Practice. John Wiley and Sons Ltd., Chichester, UK, pp. 41–77. Maurice-Bourgoin, L., Alanoca, L., Fraizy, P., Vauchel, P., 2003. Puhakka, M.E., Kalliola, R., Rajasilta, M., Salo, J., 1992. River types, Sources of mercury in surface waters of the Upper Madeira site evolution and successional vegetation patterns in erosive basins, Bolivia. Journal de Physique IV France 107, 855– Peruvian Amazonia. Journal of Biogeography 19, 651–665. 858. Revenga, C., Murray, S., Abramovitz, J., Hammond, A., 1998. Meyer, J.L., Wallace, J.B., 2001. Lost linkages and lotic ecology: Watersheds of the World: Ecological Value and Vulnerability. rediscovering small streams. In: Press, M.C., Huntly, N.J., Levin, World Resources Institute, Washington, D.C. S. (Eds.), Ecology: Achievement and Challenge. Blackwell Roux, D., de Moor, F., Cambray, J., Barber-James, H., 2002. Use of Science, Oxford, UK, pp. 295–317. landscape-level river signatures in conservation planning: a Ministe´rio das Minas e Energia, 1982. Geomorphological maps. South-African case study. Conservation Ecology 6, 6. Scale 1:1 000 000. Projeto RADAMBRASIL, Rio de Janeiro, Brasil. Sala, E., Aburto-Oropeza, O., Paredes, G., Parra, I., Barrera, J.C., Murray, S.T., Ambrose, R.F., Bohnsack, J.A., Botsford, L.W., Carr, Dayton, P.K., 2002. A general model for designing networks of M.H., Davis, G.E., Dayton, P.K., Gotshall, D., Gunderson, D.R., marine reserves. Science 298, 1991–1993. Hixon, M.A., Lubchenco, J., Mangel, M., MacCall, A., McArdle, Sanderson, E.W., Jaiteh, M., Levy, M.A., Redford, K.H., Wannebo, D.A., Ogden, J.C., Roughgarden, J., Starr, R.M., Tegner, M.J., A.V., Woolmer, G., 2002. The human footprint and the last of Yoklavich, M.M., 1999. No-take reserve networks: sustaining the wild. Bioscience 52, 891–904. fishery populations and marine ecosystems. Fisheries 24, 11– Saunders, D.L., Meeuwig, J.J., Vincent, A.C.J., 2002. Freshwater 25. protected areas: strategies for conservation. Conservation Naiman, R.J., Lonzarich, D.G., Beechie, T.J., Ralph, S.C., 1992. Biology 16, 30–41. General principles of classification and the assessment of Smith, R.K., Freemna, P.L., Higgins, J.V., Wheaton, K.S., FitzHugh, conservation potential in rivers. In: Boon, P., Calow, P., Petts, G. T.W., Ernstrom, K.J., Das, A.A., 2002. Priority areas for (Eds.), River Conservation and Management. Wiley and Sons, freshwater conservation action: a biodiversity assessment of Chichester, UK, pp. 93–123. the southeastern United States. The Nature Conservancy, Neill, C., Deegan, L.A., Thomas, S.M., Cerri, C.C., 2001. Available from: http://conserveonline.org/docs/2003/08/ Deforestation for pasture alters nitrogen and phosphorus in se_biodiv_assess.pdf. small Amazonian streams. Ecological Applications 11, 1817– Sowa, S.P., Diamond, D.D., Abbitt, R., Annis, G.M., Gordon, T., 1828. Morey, M.E., Sorensen, G.R., True, D., 2004. The Aquatic Nepstad, D., Schwartzman, S., Bamberger, B., Santilli, M., Ray, D., Component of Gap Analysis: A Missouri Prototype Final Schlesinger, P., LeFebvre, P., Alencar, A., Prinz, E., Fiske, G., Report. Missouri Resource Assessment Partnership, University Rolla, A., 2006. Inhibition of Amazon deforestation and fire by of Missouri, Columbia, Missouri. Available from: http:// parks and indigenous lands. Conservation Biology 20, 65–73. www.cerc.usgs.gov/morap/projects/aquatic_gap/ Nilsson, C., Reidy, C.A., Dynesius, M., Revenga, C., 2005. sowa_etal_dod_legacy_final_report.pdf. Fragmentation and flow regulation of the world’s large river Stein, J.L., Stein, J.A., Nix, H.A., 2002. Spatial analysis of systems. Science 308, 405–408. anthropogenic river disturbance at regional and continental Noss, R.F., Cooperrider, A.Y., 1994. Saving Nature’s Legacy: scales: identifying the wild rivers of Australia. Landscape and Protecting and Restoring Biodiversity. Island Press, Urban Planning 60, 1–25. Washington, DC, USA. Steininger, M.K., Tucker, C.J., Townshend, J.R.G., Killeen, T.J., Noss, R.F., Carroll, C., Vance-Borland, K., Wuerthner, G., 2002. A Desch, A., Bell, V., Ersts, P., 2002. Tropical deforestation in the multicriteria assessment of the irreplaceability and Bolivian Amazon. Environmental Conservation 28, 127–134. vulnerability of sites in the Greater Yellowstone Ecosystem. Thomas, J.W., Raphael, M.G., Anthony, R.G., Forsman, E.D., Conservation Biology 16, 895–908. Gunderson, A.G., Holthausen, R.S., Marcot, B.G., Reeves, G.H., Olson, D.M., Dinerstein, E., 1998. The Global 200: a representation Sedell, J.R., Solis, D.M., 1993. Viability Assessments and approach to conserving the earth’s most biologically valuable Management Considerations for Species Associated with Late- ecoregions. Conservation Biology 12, 502–515. successional and Old-growth Forests of the Pacific Northwest. BIOLOGICAL CONSERVATION 135 (2007) 500– 517 517

United States Department of Agriculture, Forest Service, and The Nature Conservancy, Arlington, Virginia. Available Washington, DC. from: http://listserv.abi.org/library/uppermsriverbasin.pdf. United States Geological Survey, 2003. Shuttle Radar Topography Wishart, M.J., Davies, B.R., 2003. Beyond catchment considerations Mission documentation: SRTM. Available from: http:// in the conservation of lotic biodiversity. Aquatic Conservation: edcftp.cr.usgs.gov/pub/data/srtm/Documentation/ Marine and Freshwater Ecosytems 13, 429–437. SRTM_Topo.txt. World Commission on Dams, 2000. Dams and Development: A Ward, J.V., 1989. The four-dimensional nature of river ecosystems. New Framework for Decision-making. Earthscan Publications Journal of the North American Benthological Society 8, 2–9. Ltd., London, UK and Sterling, VA. Weitzell, R.E., Khoury, M.L., Gagnon, P., Schreurs, B., Grossman, D., World Wildlife Fund, 2004. Biodiversity Vision of the Higgins, J., 2003. Conservation Priorities for Freshwater Southwestern Amazonian Moist Forest Ecoregion. WWF, Biodiversity in the Upper Mississippi River Basin. NatureServe Washington, DC.