Irrigation and avifaunal change in coastal Northwest Mexico: has irrigated habit attracted threatened migratory species?

Irrigation in NW Mexico

Federal investments that initiated the construction of small diversion dams and irrigation channels started at several sites of Sonora and Sinaloa in 1934, and lasted until 1970 (Ortega Noriega, 1999). Massive irrigation in coastal NW Mexico escalated after enactment of the Land and Water Reform laws of 1971 redistributed private land in the states of Sinaloa and Sonora to peasants by reducing land-holdings per owner from 100 to 20 hectares in new irrigation districts and distributing the excess land to landless families (World Bank, 1974). With these reforms, a suite of major loans from the World Bank, permitted the construction (1) of large reservoirs on most of the major rivers of coastal Sinaloa and Southern Sonora, (2) of hundreds of kilometers of cement-lined irrigation canals supplying water throughout the coastal plain, and (3) of an extensive network of drains in coastal areas that facilitated agriculture in land near sea level that had previously been too saline to farm.

Before irrigation this region received most of its water as rain that fell during a July through September monsoon (Comrie & Glenn, 1998). Now, when reservoirs are full, this region has access to irrigation water throughout the year and produces crops twice each year in some areas. The demand for water is high and some of the reservoirs in this region are so large (e.g., Huites on the Rio Fuerte) that they do not fill in the annual monsoon. Instead they are filled during typhoons and the water is stored for power and irrigation for several years. When water reserves are low, farmers are advised before the planting season that they must shift to crops demanding less water, with the result that canals are full and irrigation continues even in dry years.

Indexing faunal change

Perhaps the greatest shortcoming of specimen records is that individual records contain no information on the effort invested in obtaining them (Pyke & Ehrlich, 2010). Thus, how much time a collector actually worked in a region is not available for the vast majority of the world’s natural history specimens and we have no measures of the likelihood that a species might be encountered and collected. Although this is a major impediment to the use of specimen records in assessing changes in population densities, it can be partially overcome by using aggregated databases that can be generated through electronic catalogs such as ORNIS (http://www.ornisnet.org/home). We employ an Abundance Index (AI; AIs for multiple values) to convert simple counts of specimens to an index that is corrected by a crude measure of effort (Miki et al., 2000; Rohwer, Navarro & Voelker, 2007; Barry et al., 2009; Rohwer et al., 2012). The measure of effort is the sum of all specimens taken from the same region and time period that would be collected by the methods used to collect the focal species (Barry et al., 2009; Rohwer et al., 2012). The only attempt to evaluate how AIs preform under various sampling senarios was a demonstration that AIs computed as the percentage of adults within a single species in late summer closely paralleled the AIs for those adults computed as percentage of total passerines (Barry et al., 2009).

Formally, AI for a given species is defined as ${\displaystyle A{I}_{kr}=100\frac{x_{kr}}{\sum _{j=1}^n{x}_{jr}}}$ where x_kr is number of specimens of the kth species collected in r, the region and time period of interest, and n is the number of specimens of all species that would be “expected” to be collected using the same methods in that region and time period of interest. We say “expected” because species included in the numerator and denominator should be about equally as likely to be encountered and collected on a per individual basis. Thus, for example, species that must be collected using song playback should not be compared to species that are readily collected without playback, and nocturnal and diurnal species should not be comparable if they are not encountered and sampled at similar rates. We multiply by 100 to express AI as a percentage.

With the advent of quantitative assessments of morphological variation among populations in the mid 20th century, museums and collectors began attempting to accumulate population samples of most species they were collecting. A consequence of attempting to collect samples suitable for measuring phenotypic variability is that uncommon species that could be sampled in reasonable numbers were likely over-represented relative to their abundance in local communities, whereas common species were likely under-represented. Thus, small changes in AIs should be interpreted cautiously; nonetheless, AIs are surely better measures of faunal change than simple counts of species without correction for effort because the denominator informs us of effort (Rohwer, Navarro & Voelker, 2007; Barry et al., 2009). AIs have previously been shown to track seasonal abundance of migrants, demonstrating that they reliably track regional population changes (Rohwer et al., 2012).

Collecting methods have changed between the two time periods we are comparing. In the early collecting period most specimens were collected with shotgun, while most recent specimens were collected with mist nets. Certainly some birds are easier to collect with nets than with shotguns and vice versa but, as collectors experienced with both collecting methods, our sense is that these biases have minimal effects on total collections. Many individuals of edge and understory species that are over sampled with nets are released so that there is time to prepare good samples of species that are more difficult to collect with nets than with shotgun. Thus, the ultimate goal of acquiring a broad sample of the avian community tends to result in netted and shot birds being sampled in comparable frequencies. Although singing males are easy to collect with shotgun, our experience suggests that biases in species composition are minimal compared to biases in sex ratios generated by these alternate collecting methods. Thus in dense habitats, nets generate far more females than do shotguns, yet the number of individuals sampled per species varies less because collectors are always time-constrained. To minimize biases caused by different collecting methods we have used only cuckoos, hummingbirds, woodpeckers, and passerines to measure effort because most species in these groups are regularly collected with guns or with nets. Although we used a rifle or shotgun for some of our recent collecting, most of these new specimens were netted on joint expeditions by the University of Washington Burke Museum (UWBM) and the Museo de Zoología of the Faculty of Sciences, at the Universidad Nacional Autónoma de México (MZFC). Collecting for this project was conducted under scientific collection permits issued by the Secretaría de Medio Ambiente y Recursos Naturales (SEMARNAT) to the Facultad de Ciencias (B. Hernández-Baños FAUT 0169) of UNAM. All work was conducted in accordance with polices of the University of Washington Institutional Animal Care and Use Committee (protocol 4309-01).

The Mexican bird atlas

Since the late 1990s, A Navarro and T Peterson have been compiling a database of all the bird specimens in the world’s Natural History museums that have been collected in Mexico (Navarro et al., 2003; Peterson, Navarro-Sigüenza & Benitez-Dias, 1998). There are now more than 370,000 specimens in this database from 71 museums, including the recent collections made by the UWBM and MZFC in coastal Northwest Mexico. All of these records have been georeferenced, making it possible to use them in a GIS framework. From this master database we have extracted a set of specimens appropriate to the questions that led to this paper.

Study area

In collaboration with MZFC, the UWBM began collecting in coastal NW Mexico to study migrants that move to this region in mid to late summer to undergo their annual molt (Rohwer, Butler & Froehlich, 2005; Pyle et al., 2009). Extensive surveys revealed that most molt migrants were using agricultural fields and thorn forest habitats found below 200 m for molting (S Rohwer, 2014, unpublished data). Molt migrants apparently concentrate in this coastal region of NW Mexico for two reasons. First, the diversity and density of permanent resident species is low because these deciduous forests are mostly leafless during the November through June dry season, resulting in little competition with residents for food during the molt (Rohwer, Butler & Froehlich, 2005). Second, the July through September monsoon turns the deciduous thorn forest green, which generates an abundance of food that is exploited by molt migrants and by other migrants that breed here during the rains but winter elsewhere in Mexico.

We restrict these analyses to areas of southern Sonora and northern Sinaloa, the area of Mexico we collected and surveyed intensively for studies of molt migrants. Fortunately, Chester C. Lamb, a prolific professional collector who worked in Mexico from 1925–1955, lived in Los Mochis, Sinaloa and collected extensively in the same lowland region of coastal NW Mexico that we worked in our recent surveys. These excellent early and recent collections can be used to assess the influence of the recent development of large-scale irrigation on the avifauna of this region.

To make sampling areas comparable for early and recent collections, we eliminated from the full Mexican Atlas (1) specimens from states other than Sonora and Sinaloa, (2) specimens from months other than May–August, (3) specimens taken north of Guaymas in Sonora (27.93°N, 110.89°W) and south of the junction of Route 2 and coastal route 15 in Sinaloa (23.72°N, 106.66°W), (4) specimens collected above 200 m elevation, and (5) a set of specimens that were target-collected for particular research projects and that would have inappropriately inflated either the numerator or the denominator of the AIs. This left us with 2,100 specimens of potential use to the AI analyses. Setting the northern boundary at Guaymas, Sonora means that we include areas around Ciudad Obregón and the Rio Yaki reservoir, which supplies irrigation water to parts of southern Sonora. Most early collectors did not record elevations so the georeferenced data from the Atlas were assigned elevations using a Geographic Information System (ArcGIS 10.2; ESRI). Elevations at each collection locale (m above mean sea level) were extracted from a digital elevation model with 15 m resolution of Sonora and Sinaloa (INEGI, 2013).

Our recent collections included good numbers of birds collected in September, but we omitted September specimens from our analyses for several reasons. First, except for Yellow-billed Cuckoo (Coccyzus americanus), Rufous-winged Sparrow (Aimophila carpalis), and Varied Bunting (Passerina versicolor), most species collected in September were inappropriate to this analysis of change in the breeding avifauna of this region because most stop breeding in July or August. Second, September marks the beginning of a new influx of species that molted before they migrated south; including these birds would inflate the denominator of the AIs. Third, other local breeders, such as Tropical Kingbirds and Streak-backed Orioles, are much less common in these lowland habitats by September, apparently because most individuals of these species move to other areas in Mexico for their post-breeding molt (Pyle et al., 2009). Finally, early collectors took few September specimens in this region.

Assessing habitat changes

We used a Geographic Information System (ArcGIS 10.2) to quantify the conversion of vegetation types that occurred during the second half of the 20th century as a result of the massive expansion of irrigation. For this purpose, the study area was clipped to include only land below 200m in elevation in Northern Sinaloa and Southern Sonora (bounded by 23.72°–27.93°N, and 106.66°–110.89°W, Fig. 1). Three datasets of land use and vegetation cover, representing different time points, were obtained from the Instituto Nacional de Estadística y Geografía (INEGI, 2015, unpublished data).

Plotting AI ratios

When one component of a community is expected to increase between two time periods (our mesic species) and another is expected to remain the same or to decrease (our thorn forest species) potential changes in their AIs need to be presented in a way that scales increases and decreases in AI values equally. We have done this by computing ratios of the AI values for the pre- and post-irrigation collecting periods. When AI values for a species increased between early and late collecting periods, they are expressed as positive ratios, AI_l/AI_e, where the subscripts l and e refer to late and early collecting periods, respectively. When AI values for a species decreased between the early and late period they are plotted as negative ratios, AI_e/AI_l. Interchanging the numerator and denominator in these ratios avoids a ratio that is bounded by 0, thus scaling relative increases and decreases equivalently.

When histograms of AI ratios are used to index faunal change, species clustering near ±1 assess volatility in AI ratios that, at least in part, are driven by sampling error. This focuses attention on large changes in AI ratios, which represent species whose numbers are most likely to have changed through time. Assuming collectors attempt to assemble samples of as many species as possible, species present in sufficient numbers to be reasonably represented in two collecting periods should show little change in their AI ratios. In contrast, outliers should be species that either were so uncommon that reasonable samples could not be obtained, or that were so common that they were over-sampled when less common species were not being caught. Thus AI ratios falling beyond the peak in the distribution centered on ±1 should most reliably indicate population changes.

Land use change

The extensive development of irrigation in coastal NW Mexico has dramatically increased mesic habitat. Irrigation was taking place in this region on a local scale prior to the 1970s, but the scale of that irrigation was trivial compared to the transformation of this region following the agrarian reform laws of 1971. Irrigated agriculture on an industrial scale now accounts for the vast majority of land use on the coastal plain of northwest Mexico.

The projected natural vegetation for the region (outlined in Fig. 1) was primarily spiny thorn forest (38%) and dry scrublands (35%) as well as deciduous forest (21%). Mesic habitat was found only in areas with riparian vegetation, and comprised less than 5% of the land cover prior to irrigation. By the time of the first land-use assessment for Mexico in 1976 approximately 34% (1,293,231 hectares) of the low elevation areas of NW Mexico were categorized as mesic habitats (Fig. 1B). This was mostly due to small diversion dams and local irrigation projects and that were constructed before loans from the World Bank helped build much larger reservoirs and cement lined canals. Now, 60% of the land area (2,293,766 hectares) in our study region of coastal NW Mexico is currently covered by mesic habitat (Fig. 1C). Areas that remain unaffected by irrigation are largely hills and small mountains with slopes unsuitable for agriculture and desert habitats on the coastal plain that cannot, at present, be supplied with water.

Faunal change

Most collecting that occurred in coastal northwest Mexico prior to the development of modern irrigation occurred before the 1970s (Table 3). Then, there was a substantial to complete hiatus in collecting from the 1970s through the 1990s, during which time many major irrigation dams and canals were built. Collecting resumed with the joint UWBM/MZFC expeditions to this region conducted from 2003–2011. Of the 2100 specimens in the analysis, 1,213 were collected during the early period (Pre 1970s), predating the majority of land-use change, and 887 have been added since (Table 3).

We found no change in the rate at which the group of nine thorn forest species were collected before and after irrigation (Table 4; Fisher’s exact p = 0.28). AI values normalized by the number of species were 1.09 and 1.25, for the early and recent periods respectively (Table 4). In contrast, there was a significant increase in the normalized AI values for the group of 30 mesic species from 0.98 in the early period to 1.48 in the post-irrigation period (Table 4; Fisher’s exact p < 0.0001).

Examining changes in the AI ratios between the early and late periods of collecting (Table 2) provides further insight into the volatility of these ratios. There is a great deal more variance in the AI ratios for mesic than for thorn forest species (Fig. 2). More importantly, there is a strong right skew in the ratios for mesic species because some mesic species were seldom collected in the early period, but frequently collected in the post-irrigation period. Although collectors generally attempt to evenly sample species in a community, this skewed distribution suggests that moderate increases for a number of mesic species were, nonetheless, reflected in museum collections.

For the nine thorn forest taxa AI ratios are, as expected, little skewed (Fig. 2). Yellow Grosbeak (Pheucticus chrysopeplus) is the most negative point in this distribution, and likely has declined in total numbers at low elevations in coastal NW Mexico (Table 2; Fig. 2). Yellow Grosbeaks are never abundant in collections, and they were uncommon enough that we had little sense of their habitat requirements. We certainly over-sampled Yellow Grosbeaks, compared to their actual abundance. It was the only thorn forest species for which we collected every individual captured but densities were too low for us to capture reasonable samples.

Individual species

A few species, including two that we did not treat in the AI analyses, merit further descriptive comment and interpretation. All may have increased as a consequence of irrigation.

Conservation implications of expanded breeding numbers in Mexico

Could population changes in migrants breeding in Mexico and the US linked? When a species is winter limited (Fretwell, 1972; Terborg, 1989), the development of either higher quality habitat in NW Mexico, or of habitat similar in quality to historical breeding areas to the north that reduces migration distance, may result in a decline in numbers to the north. In contrast, if populations are regulated by density dependent breeding success, colonization of new breeding habitat in one region should not reduce breeding numbers elsewhere, as long as the winter range can accommodate summer production. Asymmetries in the land areas of breeding and wintering ranges for Neotropical migrants suggest winter limitation could be the rule; their winter range is often only a third to a fifth the size of their summer range and they arrive there with numbers swollen by summer production (Terborg, 1989).

We seldom know whether populations are primarily winter or breeding season limited; yet, most studies of threatened and endangered birds focus on losses of breeding habitat and cowbird parasitism (e.g., Goldwasser, Gaines & Wilbur, 1980; Gaines & Laymon, 1984; Kus, 1998), which implicitly assumes breeding season regulation. However, migratory shortstopping and winter population regulation can easily lead to population declines in northern breeding areas that are driven, not by losses of northern breeding habitat, but by the development of better habitat to the south. Such population shifts need not be driven solely by migratory shortstopping. Certainly, pioneering breeders in the south must be shortstoppers but, as soon as recruitment is higher in the south than in the north, increases in numbers in the south will be matched by declines in the north, if populations are winter limited.

Flexibility in the settlement patterns of Neotropical migrant passerines has been documented to occur over vast geographic scales, apparently through domino effects of territorial behavior. In Black-throated Blue Warblers (Setophaga caerulescens) adults are concentrated in the southern Appalachian Mountains where their density and productivity is high, whereas yearlings predominate to the north over large areas where densities are low (Graves, 1997). Yearling American Redstarts (Setophaga ruticilla) that depart early from their Jamaican wintering grounds settle south of their birth place in North America, but yearlings that depart late settle north of their birth place, presumably because southern habitats have filled with earlier migrants. Moreover, these distances are related to their relative departure dates (Studds, Kyser & Marra, 2008). Similarly, as measured by age ratios, the habitat distribution of Hermit and Townsend’s Warblers (Setophaga occidentalis and S. townsendi, respectively) in the Cascade Mountains of Washington and Oregon are strongly affected by elevation gradients. At low elevations adult Townsend’s Warblers show strong breeding site fidelity (Pearson, 2000), and adults at low elevations that fail to return in spring are replaced, not by yearlings, but by unmarked adults. Yearling males settle at high elevations for their first breeding season (Rohwer, 2004) but apparently moved to more suitable low-elevation sites after breeding their first season at high elevations (Pearson, 2000; Rohwer, 2004).

Most studies of the decline of migrants breeding in the American Southwest have focused almost exclusively on habitat loss in that region, probably because most of the money available for these studies must be spent in the US. This is distressing: these migrants spend the majority of the annual cycle south of the US and their population biology is little studied outside of the breeding season. If there is winter population regulation and if breeding populations in the US and coastal NW Mexico are linked by migratory shortstopping, then restored habitat in the US would need to exceed the quality of the new breeding habitat in NW Mexico sufficiently to repay the cost of a longer, trans-desert migration. The huge increase in mesic habitat, from less than 5% to 60% of the land area in coastal NW Mexico has resulted in large new breeding populations of some migrants that could be linked to declines in the US.

Eight long-distance migrants included in our assessments of changes in breeding abundance in the irrigated regions of coastal NW Mexico are reasonable candidates for such linkages (Table 5). Two of these eight species, Northern Rough-winged Swallow and Lucy’s Warbler, are molt migrants to this region (Rohwer, Navarro & Voelker, 2007; Rohwer, Hobson & Yang, 2011). For this reason the increases we estimated for these species may be high, so we do not further consider these species. Two others, Yellow-breasted Chat (Icteria virens) and Common Yellowthroat (Geothlypis trichas), appear to have changed little in abundance in coastal NW Mexico and neither shows BBS (Sauer et al., 2013) declines in the west (Table 5). Chats have likely expanded with the increase in brushy edges, but they are well represented in collections prior to the 1970s; our recent collections seldom included breeding yellowthroats, which were common only along the coast (Table 2). In contrast, Yellow-billed Cuckoo, Cliff Swallow, Bell’s Vireo and Orchard Oriole, have likely increased dramatically as breeders in the irrigated regions of NW Mexico. All are migrants with breeding ranges mostly in the US and Canada and all have declined to the north, suggesting their population changes could be linked.

Yellow-billed Cuckoos have disappeared from most of their historic breeding range in the western US and Canada (Goldwasser, Gaines & Wilbur, 1980; Gaines & Laymon, 1984; Laymon & Halterman, 1987). Although much riparian habitat has been lost in the Southwest US, the loss of cuckoos from northern parts of their western breeding range is inconsistent with their decline being driven by losses of breeding habitat because landscape changes in the north have increased ecotonal habitats they prefer for breeding. Although historically widely distributed in the west, the western population of cuckoos was probably never very large because they were restricted to riparian corridors and human created ecotones. The density of cuckoos now breeding in the irrigated habitats of coastal NW Mexico easily seems high enough to account for the numbers that previously bred in the western US and Canada. Further, the decline in the much larger population of cuckoos in eastern North America (Sauer et al., 2013), suggests that loss of winter habitat may explain their continent-wide decline in breeding numbers, even though they surely have increased as breeding birds in coastal NW Mexico.

For the endangered Least’s Bell’s Vireo (Vireo bellii pusillus) of California the situation is less compelling, but shortstopping seems worth considering given the remarkable abundance of Bell’s Vireos that now breed in irrigated regions of NW Mexico. The California population has responded positively to habitat improvements and to cowbird control, which weakens a case for a causal link between the expanding population in NW Mexico and declines in California (Kus, 1998; Kus & Whitfield, 2005). Pusillus supposedly winters only in Baja California, but a small change in the fall migratory orientation of pusillus that breed in California would take them to the excellent new breeding habitat created by irrigation in coastal Sonora and Sinaloa.

With an average of almost $2.5 m spent on recovery efforts each year for Least Bell’s Vireos (Restani & Marzluff, 2001), it seems important to assess whether their decline in the US might be linked to their increase in NW Mexico. To assess the source of the Bell’s Vireos now breeding in Sinaloa, we attempted to assign 25 new Sineloa specimens to race: 18 better matched V.b. arizonae, but 7 better matched V.b. pusillus. Distinguishing these races is difficult, so this conclusion should be further evaluated using genetics or geolocators (Klicka, 2014). If further genetic work using SNPs (Klicka, 2014) discloses pusillus genotypes in the now large breeding populations in NW Mexico, then habitat restoration in California could be competing with the recently developed irrigated habitat in coastal Sonora and Sinaloa. Alternatively, if the V. bellii now breeding in coastal NW Mexico all carry arizonae SNPs, then there would be no reason to think shortstopping in Sinaloa and Sonora has contributed to the decline of pusillus in California.

Unlike cuckoos and vireos, breeding Cliff Swallows remain abundant in Canada and the western USA, but declines have been substantial since the Breeding Bird Survey was begun. Our AIs greatly underestimate the increase in Cliff Swallows, and the numbers we see breeding in NW Mexico easily seem sufficient to account for their decline in Canada and the western US.

Finally, Orchard Orioles were not previously thought to breed in NW Mexico (Miller et al., 1959). Not only are they common breeders in irrigated farmlands of this region (Rohwer & Wood, 2013), but their breeding densities around irrigated alfalfa fields similar to those found in the best breeding habitats in Kansas and Louisiana. The BBS shows significant survey-wide declines for Orchard Orioles, but their sharpest declines have been in southern states (Table 5). Because they are patchily distributed as breeding birds in Mexico and in the USA, we have little sense of whether the numbers now breeding in the irrigated habitats of NW Mexico are sufficient to account for the declines in breeding numbers to the north.

Conclusions

Widespread irrigation has enormously increased the percentage of mesic land cover in coastal NW Mexico, leaving less than 40% of the land area in dry habitat. This change has increased the diversity and abundance of small birds breeding in this region, without resulting in the loss of species dependent on the original thorn forest habitat. The greatest value of discovering new breeding populations or large increases in breeding numbers of some migrants in this region is the implication that these population increases of Mexico could be linked, through winter population regulation, to declines in breeding numbers in the US and Canada. Given the sums being invested in these populations in the US (Restani & Marzluff, 2001), further exploring this possible linkage seems urgent. In general, both the connections between events affecting winter and summer ecology of migrants (Wilcove & Wikelski, 2008) and the degree to which habitat distributions may be affected through domino effects of social behavior over landscape and, possibly, range-wide scales (Graves, 1997; Rohwer, 2004; Studds, Kyser & Marra, 2008) seems underappreciated. If breeding numbers to the north are causally related to the development of new breeding populations in Mexico, then recovery investments for those species in the US and Canada should be redirected to other management priorities.