Using Imagery to Monitor Riparian and Upland Vegetation Along the San Pedro River, Arizona
The Upper San Pedro River is one of the few remaining undammed rivers that maintain a vibrant riparian ecosystem in the southwest. However, its riparian forest is threatened by diminishing groundwater and surface water inputs, due to either changes in watershed characteristics such as changes in riparian and upland vegetation, or human activities such as regional groundwater pumping. We used Landsat normalized difference vegetation index (NDVI) to quantify the green leaf density of the groundwater-dependent forest from 1984 to 2012. Pre-monsoon NDVI values showed a 20% drop for the northern reach and no net change for the southern reach. NDVI was positively correlated with river flows, which decreased in the northern reach, and negatively correlated with air temperatures in both reaches, which have increased by 1-4 °C from 1932 to 2012. NDVI in the uplands did not increase from 1984 to 2012, suggesting that increased water use was not a factor in reducing river flows. Climate change, regional groundwater pumping, changes in the intensity of monsoon rain events and lack of overbank flooding are feasible explanations for deterioration of the riparian forest in the northern reach.
Background & Importance
Human activities and climate change have negatively impacted riparian forests throughout the world’s arid and semiarid zones. However, it is difficult to document long-term consequences of land use and climate change and to pinpoint environmental drivers of deterioration. Many cases of deterioration are due to direct impacts on river systems such as diversion of water for human use, flow regulation and introduction of invasive species. However, even unregulated and protected rivers can be impacted by regional and global changes in hydrology and climate.
A case in point is the Upper San Pedro River in northwestern Mexico and southeastern Arizona, United States. This river originates in Mexico and flows north into the United States, ultimately discharging into the Gila River in the Lower Colorado River Basin. Much of the riparian zone is protected within the San Pedro Riparian National Conservation Area (SPRNCA), a 55 km long portion of the river. Created in 1988, the SPRNCA is the United States’ only designated Riparian National Conservation Area. Agriculture, groundwater extraction and livestock grazing have been eliminated within the SPRNCA. However, regional groundwater pumping to support population growth and possible changes in climate in the area have raised concerns about the health of the riparian forest along the river.
Numerous organized research efforts have been undertaken over the last couple of decades with the goal of determining the trajectory of ecological changes in the SPRNCA in response to land cover and land use changes, hydrological factors and potential effects of climate change. The scope of the studies encompassed the riparian zone itself as well as the tributary streams and the sparsely vegetated uplands. The San Pedro is one of the best-studied rivers in the western United States. Nevertheless, most of the studies have been conducted over a relatively short period of time, and the longer-term trends in riparian vegetation cover and their relationship to meteorological and hydrological factors in the Upper San Pedro Basin are still in dispute.
The goal of this study was to document long-term changes in green vegetation density in both the riparian corridor and in the surrounding uplands through satellite imagery, to shed light on the status of the riparian ecosystem and upland vegetation density that might be related to diminished flows. The specific objectives were as follows: (1) determine the multi-decadal trajectory of riparian and upland vegetation along the Upper San Pedro River based on satellite vegetation index values and (2) correlate vegetation index changes with concurrent hydrological and meteorological changes in the watershed.
General Methods
The study employed data from the Landsat Thematic Mapper (TM) satellites and from the Moderate Resolution Imaging Spectrometer (MODIS) sensors on the Terra satellite. Landsat 5, the primary TM satellite used in this study, has provided coverage since 1984 at 30m resolution, sufficient for mapping individual vegetation units. MODIS has provided coverage since 2000 at 250m resolution but has near-daily coverage, so temporal coverage is much better than Landsat with a 16-day return time. Annual changes in NDVI and enhanced vegetation index (EVI) values in the riparian corridor were compared with changes in depth to water table (DTW), river flows (Flows), air temperature (Tair) and PPT to develop predictive models of the vegetation response to long-term trends in environmental variables. We divided the river into two reaches, one downstream (north) and one upstream (south) of the USGS streamflow-gaging station near Charleston, AZ, where a decrease in flows has been detected. We also determined NDVI and EVI values for the upland areas in the watershed to see if decreased river flows are correlated with increased green vegetation density and therefore ET in the uplands.
Important Results
Landsat NDVI and EVI in the riparian zone, 1984–2012
Shape files based on the 1984 TM image showed that the riparian corridor covered 3951 ha, of which 2791 ha was riparian habitat and 1160 ha were agricultural fields within the riparian corridor. Both NDVI and EVI decreased significantly over time in the riparian areas in the north reach. After subtracting out the approximate value of bare soil (NDVI = 0.17 based on sampling non-vegetated areas within the riparian corridor), the mean NDVI for riparian vegetation decreased by about 20% from 1984 to 2012 (0.7% per year) (r2 = 0·37). NDVI did not decrease significantly in the south reach; EVI decreased significantly but with lower slope than the decrease for EVI in the north reach. Vegetation indices of agricultural fields decreased by about 50% over the same time period (r2 = 0·70). Most of the fields were still visible in the images, but they had much lower NDVI values than in the 1984 image, presumably because they were no longer cultivated.
NDVI differences maps
Normalized difference vegetation index difference maps comparing 1984 and 2010 are shown in the published manuscript for the north and the south reach. For combined agriculture and non-agriculture classes in the riparian corridor, 926 ha were red (decreased NDVI) while 280 ha were green (increased NDVI). For the agriculture class, 737 ha were red and 181 ha were green, supporting the NDVI time-series showing retirement of fields. Most of the retired fields were in the south reach of the river, whereas some new fields (green pixels) appeared in the north reach. For the non-agriculture classes in the riparian zone, 189 ha were red and 99 ha were green. Thus, the pixel-by-pixel analysis qualitatively supports the NDVI trends, with both classes decreasing but with agriculture decreasing the most. Red pixels accounted for 9·94% of total pixels in the north reach and 6·75% in the south reach, a difference that was not significant. However, green pixels accounted for only 0·726% of total pixels in the north reach and 2·71% in the south reach. The results reinforce the time-series data showing a long-term net decline in riparian NDVI, confined mainly to the north reach.
Examples of vegetation changes over time
Landsat imagery did not have sufficient resolution to determine which species increased or decreased. However, by comparing high-resolution archival imagery with Landsat NDVI difference maps, it was possible to qualitatively evaluate the nature of vegetation changes. Examination of archival high-resolution imagery showed that individual vegetation units were remarkably stable over time. Over the whole river, for example, most of the cottonwood trees visible on archival images dating from 1992 were still present in 2013. Examples of vegetation changes are in shown in the publication, comparing archival imagery with NDVI difference images. At Site 1 in middle of the north reach, cottonwoods fringing the river channel showed a net decrease in NDVI, apparently because of the loss of individual trees over time, based on archival imagery. At Site 2 further upstream (i.e., south) in the north reach, cottonwood trees showed an increase in NDVI, because of growth of tree canopies, whereas mesquite and grass stands away from the main channel decreased. On the other hand, at Site 3 further upstream in the south reach, cottonwood trees along the main channel showed a clear increase in growth and NDVI, while several trees away from the main channel apparently died by 2010.
Future Directions
Causes and consequences of future decreases in river flows
Relationships between river flows, groundwater and riparian NDVI and EVI suggest that decreasing surface flows and rates of alluvial aquifer recharge is already impacting the riparian forest, at least during the pre-monsoon period, especially in the northern reach. As noted by Thomas and Pool (2006), agricultural pumping is a relatively small term in the water budget, and agricultural NDVI decreased by 50% from 1984 to 2012, so it cannot explain the magnitude of the reduction in river flows or riparian NDVI, or the increase in DTW. Changes in vegetation do not seem to be sufficient to explain decreases in San Pedro River flows at least from 1984 to the present, and the trend towards less-intense rainfall events did not extend beyond 1998 (Goodrich et al., 2008), but decreases in river flows have continued. The cause or causes of the reduction in river flows remain uncertain (Thomas and Pool, 2006; Stromberg et al., 2009b; Goodrich et al., 2011), but future research should continue to focus on the relationship between regional pumping, flows in the river and the health of the riparian forest in SPRNCA. The present study shows that a slow decline in riparian vegetation, predicted by Stromberg et al. (2009a) based on flow reductions and lack of recruitment of new cohorts of phreatophytes, is already in progress.
The Upper San Pedro River is one of the few remaining undammed rivers that maintain a vibrant riparian ecosystem in the southwest. However, its riparian forest is threatened by diminishing groundwater and surface water inputs, due to either changes in watershed characteristics such as changes in riparian and upland vegetation, or human activities such as regional groundwater pumping. We used Landsat normalized difference vegetation index (NDVI) to quantify the green leaf density of the groundwater-dependent forest from 1984 to 2012. Pre-monsoon NDVI values showed a 20% drop for the northern reach and no net change for the southern reach. NDVI was positively correlated with river flows, which decreased in the northern reach, and negatively correlated with air temperatures in both reaches, which have increased by 1-4 °C from 1932 to 2012. NDVI in the uplands did not increase from 1984 to 2012, suggesting that increased water use was not a factor in reducing river flows. Climate change, regional groundwater pumping, changes in the intensity of monsoon rain events and lack of overbank flooding are feasible explanations for deterioration of the riparian forest in the northern reach.
Background & Importance
Human activities and climate change have negatively impacted riparian forests throughout the world’s arid and semiarid zones. However, it is difficult to document long-term consequences of land use and climate change and to pinpoint environmental drivers of deterioration. Many cases of deterioration are due to direct impacts on river systems such as diversion of water for human use, flow regulation and introduction of invasive species. However, even unregulated and protected rivers can be impacted by regional and global changes in hydrology and climate.
A case in point is the Upper San Pedro River in northwestern Mexico and southeastern Arizona, United States. This river originates in Mexico and flows north into the United States, ultimately discharging into the Gila River in the Lower Colorado River Basin. Much of the riparian zone is protected within the San Pedro Riparian National Conservation Area (SPRNCA), a 55 km long portion of the river. Created in 1988, the SPRNCA is the United States’ only designated Riparian National Conservation Area. Agriculture, groundwater extraction and livestock grazing have been eliminated within the SPRNCA. However, regional groundwater pumping to support population growth and possible changes in climate in the area have raised concerns about the health of the riparian forest along the river.
Numerous organized research efforts have been undertaken over the last couple of decades with the goal of determining the trajectory of ecological changes in the SPRNCA in response to land cover and land use changes, hydrological factors and potential effects of climate change. The scope of the studies encompassed the riparian zone itself as well as the tributary streams and the sparsely vegetated uplands. The San Pedro is one of the best-studied rivers in the western United States. Nevertheless, most of the studies have been conducted over a relatively short period of time, and the longer-term trends in riparian vegetation cover and their relationship to meteorological and hydrological factors in the Upper San Pedro Basin are still in dispute.
The goal of this study was to document long-term changes in green vegetation density in both the riparian corridor and in the surrounding uplands through satellite imagery, to shed light on the status of the riparian ecosystem and upland vegetation density that might be related to diminished flows. The specific objectives were as follows: (1) determine the multi-decadal trajectory of riparian and upland vegetation along the Upper San Pedro River based on satellite vegetation index values and (2) correlate vegetation index changes with concurrent hydrological and meteorological changes in the watershed.
General Methods
The study employed data from the Landsat Thematic Mapper (TM) satellites and from the Moderate Resolution Imaging Spectrometer (MODIS) sensors on the Terra satellite. Landsat 5, the primary TM satellite used in this study, has provided coverage since 1984 at 30m resolution, sufficient for mapping individual vegetation units. MODIS has provided coverage since 2000 at 250m resolution but has near-daily coverage, so temporal coverage is much better than Landsat with a 16-day return time. Annual changes in NDVI and enhanced vegetation index (EVI) values in the riparian corridor were compared with changes in depth to water table (DTW), river flows (Flows), air temperature (Tair) and PPT to develop predictive models of the vegetation response to long-term trends in environmental variables. We divided the river into two reaches, one downstream (north) and one upstream (south) of the USGS streamflow-gaging station near Charleston, AZ, where a decrease in flows has been detected. We also determined NDVI and EVI values for the upland areas in the watershed to see if decreased river flows are correlated with increased green vegetation density and therefore ET in the uplands.
Important Results
Landsat NDVI and EVI in the riparian zone, 1984–2012
Shape files based on the 1984 TM image showed that the riparian corridor covered 3951 ha, of which 2791 ha was riparian habitat and 1160 ha were agricultural fields within the riparian corridor. Both NDVI and EVI decreased significantly over time in the riparian areas in the north reach. After subtracting out the approximate value of bare soil (NDVI = 0.17 based on sampling non-vegetated areas within the riparian corridor), the mean NDVI for riparian vegetation decreased by about 20% from 1984 to 2012 (0.7% per year) (r2 = 0·37). NDVI did not decrease significantly in the south reach; EVI decreased significantly but with lower slope than the decrease for EVI in the north reach. Vegetation indices of agricultural fields decreased by about 50% over the same time period (r2 = 0·70). Most of the fields were still visible in the images, but they had much lower NDVI values than in the 1984 image, presumably because they were no longer cultivated.
NDVI differences maps
Normalized difference vegetation index difference maps comparing 1984 and 2010 are shown in the published manuscript for the north and the south reach. For combined agriculture and non-agriculture classes in the riparian corridor, 926 ha were red (decreased NDVI) while 280 ha were green (increased NDVI). For the agriculture class, 737 ha were red and 181 ha were green, supporting the NDVI time-series showing retirement of fields. Most of the retired fields were in the south reach of the river, whereas some new fields (green pixels) appeared in the north reach. For the non-agriculture classes in the riparian zone, 189 ha were red and 99 ha were green. Thus, the pixel-by-pixel analysis qualitatively supports the NDVI trends, with both classes decreasing but with agriculture decreasing the most. Red pixels accounted for 9·94% of total pixels in the north reach and 6·75% in the south reach, a difference that was not significant. However, green pixels accounted for only 0·726% of total pixels in the north reach and 2·71% in the south reach. The results reinforce the time-series data showing a long-term net decline in riparian NDVI, confined mainly to the north reach.
Examples of vegetation changes over time
Landsat imagery did not have sufficient resolution to determine which species increased or decreased. However, by comparing high-resolution archival imagery with Landsat NDVI difference maps, it was possible to qualitatively evaluate the nature of vegetation changes. Examination of archival high-resolution imagery showed that individual vegetation units were remarkably stable over time. Over the whole river, for example, most of the cottonwood trees visible on archival images dating from 1992 were still present in 2013. Examples of vegetation changes are in shown in the publication, comparing archival imagery with NDVI difference images. At Site 1 in middle of the north reach, cottonwoods fringing the river channel showed a net decrease in NDVI, apparently because of the loss of individual trees over time, based on archival imagery. At Site 2 further upstream (i.e., south) in the north reach, cottonwood trees showed an increase in NDVI, because of growth of tree canopies, whereas mesquite and grass stands away from the main channel decreased. On the other hand, at Site 3 further upstream in the south reach, cottonwood trees along the main channel showed a clear increase in growth and NDVI, while several trees away from the main channel apparently died by 2010.
Future Directions
Causes and consequences of future decreases in river flows
Relationships between river flows, groundwater and riparian NDVI and EVI suggest that decreasing surface flows and rates of alluvial aquifer recharge is already impacting the riparian forest, at least during the pre-monsoon period, especially in the northern reach. As noted by Thomas and Pool (2006), agricultural pumping is a relatively small term in the water budget, and agricultural NDVI decreased by 50% from 1984 to 2012, so it cannot explain the magnitude of the reduction in river flows or riparian NDVI, or the increase in DTW. Changes in vegetation do not seem to be sufficient to explain decreases in San Pedro River flows at least from 1984 to the present, and the trend towards less-intense rainfall events did not extend beyond 1998 (Goodrich et al., 2008), but decreases in river flows have continued. The cause or causes of the reduction in river flows remain uncertain (Thomas and Pool, 2006; Stromberg et al., 2009b; Goodrich et al., 2011), but future research should continue to focus on the relationship between regional pumping, flows in the river and the health of the riparian forest in SPRNCA. The present study shows that a slow decline in riparian vegetation, predicted by Stromberg et al. (2009a) based on flow reductions and lack of recruitment of new cohorts of phreatophytes, is already in progress.