longitudinal study of the impacts of land cover change on ......55" some of the very earliest...
TRANSCRIPT
1
Longitudinal Study of the Impacts of Land Cover Change on Peak Flows in Four Mid-sized 1 Watersheds in New York State, USA 2
3
Stephen B. Shaw*, John Marrs, Nishan Bhattarai, and Lindi Quackenbush 4
SUNY College of Environmental Science and Forestry, Dept. of Environmental Resources Engineering, 5
Syracuse, NY 13210 6
*contact info - phone: 315-470-6939; email: [email protected] 7
8
2
Abstract 9
In humid, temperate regions, there remains limited direct evidence of the influence of land cover changes 10
on hydrologic response (e.g. peak discharge), especially across larger watersheds. Using historic aerial 11
photography in conjunction with long-term stream gaging data, we assessed the role of land cover change 12
on hydrologic response over multi-decadal periods in four watersheds in New York State. All four 13
watersheds had increases in forest cover accompanied by small increases in urban land cover. Hydrologic 14
response was evaluated by establishing an empirical function relating precipitation, watershed wetness, 15
and discharge for each era of distinct land cover. This function was then used to estimate discharge for 16
fixed precipitation amounts and wetness levels, allowing weather variables to be controlled across eras. 17
One watershed (Limestone Creek) exhibited virtually no change in hydrologic response despite forest 18
cover increasing by over 100%. One watershed (Fall Creek) exhibited a slight increase in hydrologic 19
response with a greater than 100% increase in forest cover. The two other watersheds exhibited a greater 20
than 20% decrease in hydrologic response, but we speculate the changes in these two watersheds were in 21
part due to the construction of numerous small dams (Wappinger Creek) and a possible loss of riparian 22
wetlands (Sterling Creek). This work demonstrates that the effects of land cover on hydrologic response 23
are not always consistent with standard hydrologic intuition (i.e. increasing forested land does not always 24
reduce peak discharge) and that often other factors may be more important than basic land cover in 25
controlling hydrologic response. 26
27
28
Keywords: land cover change, dams, hydrology, floods 29
30
31
3
32
1. Introduction 33
Engineering hydrology has long made use of modeling frameworks that strongly emphasize the 34
sensitivity of runoff production to aggregate land cover. In particular, nearly all engineering hydrology 35
textbooks introduce the Curve Number method (USDA SCS 1972) as a simple tool to model rainfall-36
runoff relationships. The Curve Number equation is fundamentally a one-parameter model in which the 37
single model parameter is dependent in part on soil type but mainly on land cover. For instance, based on 38
the Curve Number method, a transition from forested to row crop agriculture in a watershed would result 39
in a 50% increase of runoff for a 51 mm precipitation event. Besides its direct use as an event-based 40
rainfall-runoff tool, the Curve Number method is embedded in commonly used computational hydrologic 41
models such as SWAT (Arnold et al. 1990) and AGNPS (Young et al. 1994). One could reasonably say 42
that the notion that hydrologic response is closely tied to land cover is deeply ingrained in engineering 43
hydrology. 44
45
Despite the broad acceptance of the Curve Number method and the accompanying implicit assumption of 46
the dominant control of land cover on hydrologic response, direct experimental measurements remain 47
limited in their ability to quantify the sensitivity of hydrologic response to land cover change. These 48
limitations arise from four sources: 1. mismatches in scale between experiments and application, 2. 49
difficulty in separating weather variability from variations due to land cover, 3. difficulty in controlling 50
for watershed features besides land cover, and 4. a focus of many studies on drastic changes in land cover 51
(forest to urban; forest to clear cut) but not on more moderate changes in land use. Below we briefly 52
discuss how each of these four limitations have been addressed in different experimental designs. 53
54
4
Some of the very earliest hydrology experiments focused on discerning the role of land cover on 55
hydrologic response (e.g. Bates and Henry 1928). These experiments took the form of “paired” watershed 56
studies in which two nearby watersheds were selected and then land cover in one of the basins was 57
modified, typically by removing trees. Discharge in the paired basins would be correlated prior to 58
disturbance to establish a baseline and then correlation between post-disturbance discharge would be 59
compared to the baseline measurements. These paired watershed studies are ideal to control for non-land 60
cover features that may regulate hydrologic response as well as for controlling for weather variability. 61
However, to allow for direct modification of the watershed they are often conducted on watersheds less 62
than 2 km2 in area (Andreassian 2004) and often only entail relatively dramatic transitions in land cover 63
such as from fully vegetated conditions to clear cut conditions. 64
65
For larger watersheds that better represent the heterogeneities of most basins, studies typically follow a 66
“comparative” approach where multiple basins of varying land cover are compared to each other over the 67
same period of time (e.g. Rose and Peters 2001, Poff et al. 2006). In this case, there is limited control on 68
non-land cover features, but weather variability can be controlled for as long as basins are evaluated over 69
the same time period and receive similar storm events. Non-land cover features can include such things as 70
geology, soils, and topography. 71
72
A third option is to conduct a “longitudinal” study in which the same basin is analyzed over time. A 73
longitudinal study requires that the basin has land cover change over the period in which it is gaged. 74
Because of the often gradual change in land cover, such basins need to be monitored over many decades. 75
There are relatively few examples of longitudinal studies in the literature (e.g. Beighley and Moglen 76
2002). Where comparative studies are limited by their ability to control for non-land cover features, 77
longitudinal studies are limited by their ability to control for weather. Arguably though, it is easier to 78
5
control for weather than to control for innate watershed features. By normalizing precipitation input 79
within the same time scale of the hydrologic response being assessed (e.g. calculating runoff ratio or some 80
variant), one can reasonably account for variations in weather drivers over time (Sriwongsitanon and 81
Taesombat 2011, Beighley and Moglen 2002). 82
83
This paper uses a longitudinal approach to evaluate sensitivity to land cover change, specifically focusing 84
on the transition from agricultural to forested land in New York State. Most North American studies to 85
date focus on differences between urban and non-urban land cover (Burns et al. 2004, Schoonover et al. 86
2006, Boggs and Sun 2011). While there are recent studies in developing countries (predominantly in 87
tropical regions in the southern hemisphere) that have looked at the transition from forest to agricultural 88
land use (Bradshaw et al. 2011, Bathhurst et al. 2007, Sriwongsitanon and Taesombat 2011), most of 89
these studies are not necessarily representative of the processes driving hydrological response in humid, 90
temperate regions in northern latitudes. 91
92
In this study, we make use of an historical aerial photography dataset dating back to the 1930’s that 93
covers relatively large swaths of New York State. By using the same watershed but looking at long-term 94
changes, we remove the possible influence of non-land cover features. By controlling for precipitation 95
input and antecedent moisture, we account for long-term variations in weather. The study focuses on 96
moderate sized watersheds with moderate land cover change (transition on approximately 25% of 97
watershed area) thus filling in a knowledge gap regarding moderate, but not drastic, land cover change. 98
99
2. Methods 100
6
The overarching objective of this study is to match hydrologic response to land cover statistics in 101
different eras for individual watersheds to assess whether land cover changes translate to changes in 102
hydrologic response. There are two primary methodological tasks: 1. compiling geo-referenced land cover 103
maps and 2. calculating the hydrologic response metric for each era. 104
105
We analyzed four mid-sized watersheds in New York State. Since the early 1900’s, large tracts of farm 106
land in rural parts of New York have been abandoned and allowed to revert back to forest. The 107
watersheds were selected based on availability of long-term USGS stream gaging records, availability of 108
precipitation data from the National Weather Service, and availability of historic aerial imagery. Mid-109
sized watersheds on the order of 100 km2 were selected in order to reduce heterogeneity in precipitation 110
inputs across the watershed, to not require overwhelming effort to accurately classify land cover change, 111
and to minimize the importance of in-channel travel time in assessing hydrologic response. In terms of in-112
channel travel times, the daily precipitation is the cumulative flux reported in the early-morning on the 113
reporting day while daily streamflow is the 24-hour average extending from midnight to midnight on that 114
same reporting day. Thus, the difference in reporting protocols for streamflow and precipitation innately 115
account for travel time delays on the order of hours. 116
117
2.1 Watershed Characteristics 118
Watershed locations are indicated in Figure 1. Fall Creek and Limestone Creek are situated within the 119
Appalachian Plateau physiographic region of New York. Soils in the Appalachian Uplands tend to consist 120
of glacial till on hillslopes and glacial fluvial deposits (sand and gravel outwash) in the valley bottoms 121
(Lumia 1991). The northern portion of the Limestone Creek watershed intersects an outcropping of 122
Devonian limestone. This formation dips to the south, so much of the subsurface of the watershed is 123
presumably underlain by this limestone formation. There have been no geohydrological studies of 124
7
surface-subsurface interactions within the Limestone Creek watershed, but studies in karst geology in the 125
same formation further east in New York State indicate the possible presence of extensive subsurface 126
karst drainage networks that interact with more surficial flow paths (Baker 1973). Wappinger Creek is 127
located in the Appalachian Valley and Ridge province. The watershed primarily overlays sedimentary 128
bedrock, but there are zones of metamorphic rock in outcroppings. Extensive deposits of sand and gravel 129
can be found in the valley bottoms. In general, the Limestone, Fall, and Wappinger Creek watersheds are 130
relatively similar, consisting of shallow soils overlaying shale (with some limestone and dolomite) on 131
moderate to gently sloping terrain (Lumia 1991). The geology and soils of the Sterling Creek watershed 132
are somewhat different. It is located in the central lowlands (lake plain) province of New York (Lumia 133
1991). Soils in the central lowlands are primarily glaciolucustrine in origin and tend to be poorly drained 134
with a high water table. The terrain is primarily flat with the exception of several drumlins in the southern 135
portion of the watershed. Basic watershed characteristics are summarized in Table 1. 136
137
Besides differences in innate physical features, we noted a distinct difference in the degree of 138
anthropogenic hydrologic modification across the study sites. In particular, the Wappinger Creek 139
watershed has a large number of small impoundments created by small dams. Based on the New York 140
State Department of Environmental Conservation (NYS DEC) dam inventory available through the New 141
York State GIS Clearinghouse (http://gis.ny.gov/gisdata/inventories/member.cfm?organizationID=529; 142
last accessed 11/25/2013), Wappinger Creek has 85 dams within its boundaries, while Fall Creek has 29, 143
Limestone Creek has 14, and Sterling Creek has seven. Wappinger Creek has more than twice as many 144
dams per unit watershed area than the other three watersheds. Additionally, 50% of the dams in 145
Wappinger Creek appear to have been built between 1950 and 1990 with many of the larger ones built 146
during this time, directly overlapping with the period of land cover transition we evaluated. While other 147
watersheds also had some dam building between 1950 and 1990, the largest dams tended to be built 148
earlier or later. We will consider the possible influence of these dams in more depth later in the paper. 149
8
150
2.2 Land Cover Mapping 151
Sterling Creek watershed land cover was digitized as part of a prior research project; geo-referenced 152
raster maps of land cover for the watershed were available from the Cornell University Geospatial 153
Repository (CUGIR) for the 1950’s, 1980’s, and 2000’s. For the Fall Creek and Limestone Creek 154
watersheds, non-digitized historic aerial photographs from 1930s and 1960s were obtained from the 155
Cornell Institute for Resource Information Systems (IRIS) Aerial Photograph Collection (http://aerial-156
ny.library.cornell.edu/; last accessed 11/25/2013). These photographs were panchromatic, direct contact 157
prints (i.e. printed directly from negative to paper with no enlargement or reduction) taken vertically 158
using aerial mapping cameras. A total of 108 and 55 photographs were obtained for deriving land cover 159
maps of Fall Creek and Limestone Creek, respectively. The photographs were registered to 2006 digital 160
orthoimagery (New York State Digital Orthoimagery Program; http://gis.ny.gov/gateway/mg/; last 161
accessed 11/25/2013) using a well distributed group of at least 10 reference points representing road 162
intersections common to both images. Georeferenced aerial photographs for the 1930’s for the Wappinger 163
Creek watershed were obtained directly from the New York State GIS Clearinghouse. 164
165
The primary goal of the land cover classification was to identify three classes—urban, agriculture, and 166
forested regions—within the Fall Creek, Limestone Creek, and Wappinger Creek watersheds. Because of 167
the low spectral resolution in the available imagery, experimentation showed that a two-step classification 168
procedure was required. This experimentation also suggested that there was no advantage to using 169
advanced classifiers, thus an unsupervised classification was used. The first step consisted of performing 170
a preliminary image classification to identify forest and transitional land cover from each geo-referenced 171
photograph using the Iso Cluster tool in the ArcGIS 10 software package (Esri 2011). Transitional areas 172
were identified with sparsely distributed shrub or tree-like vegetation and were assumed to be abandoned 173
agricultural land in a period of successional growth. These transitional areas were included in the tally of 174
9
forested land cover. This preliminary image was then edited manually to identify urban areas and correct 175
any obvious inconsistencies in the classification of forest and transitional areas. Figure 2 shows a 176
representative example of an aerial image and the area classified as forest land cover. Any areas not 177
identified as urban or forest were assumed to be agricultural land. Individual classified images were 178
combined into a watershed-wide layer using the mosaic function of ERDAS IMAGINE (ERDAS Inc., 179
Norcross, GA) software. Land cover maps for the 1990s were obtained from the 1992 National Land 180
Cover Dataset (Vogelmann et al. 2001). Land cover for each watershed in each era is summarized in 181
Table 2. 182
183
2.3 Hydrologic Response Metrics 184
Stream discharge is a direct outcome of hydroclimatological controls. If there are few large precipitation 185
events in a given era, then there will be few large discharge events. Thus, it is understood that a key 186
requirement of a longitudinal study of land cover change is to control for variations in precipitation. Past 187
studies (Sriwongsitanon and Taesombat 2011, Beighley and Moglen 2002) that have controlled for 188
precipitation inputs have calculated runoff ratios, the ratio of discharge (normalized by watershed area) to 189
precipitation. However, there is also recent recognition that peak stream discharges are not dictated by 190
event precipitation alone but are also dependent on antecedent water storage across a catchment (Van 191
Steenbergen and Willems 2013, Merz and Blӧschl 2009, Shaw and Walter 2009). Thus, we control for 192
both event precipitation amount and antecedent conditions prior to peak discharge events. 193
194
Instead of using the runoff ratio, we use a related metric that, like the runoff ratio, is also calculated only 195
from observed area normalized discharge (Q) and precipitation (P): 196
PQPS −=2
Eqn. 1 197
10
where S is the conceptual available watershed moisture storage. The advantage of this metric over the 198
standard runoff ratio is that it directly accounts for non-linearity in the rainfall-runoff relationship (i.e. 199
higher precipitation amounts will result in disproportionately higher discharge). Notably, this metric 200
moves inversely to the runoff ratio; a high runoff ratio indicates a low S, as would be expected given that 201
S is representative of available watershed moisture storage. Equation 1 originates from algebraic 202
manipulation of the Curve Number equation. We hesitate to introduce Equation 1 as the Curve Number 203
equation since it does not make use of the tabulated Curve Number values that are the defining feature of 204
the Curve Number equation in practice. When solving directly for Q, we consider Equation 1 to be a 205
simple, one-parameter rainfall-runoff model in the same sense that the runoff ratio can be used as a simple 206
one-parameter rainfall-runoff model (Q=Runoff Ratio × P). 207
208
Using Equation 1, we calculate event S values for upwards of 30 different storm events in each era within 209
each watershed. Storm events are selected so as to have spatially uniform precipitation across the basin 210
and be relatively temporally isolated such that there is an obvious impulse-response relationship between 211
observed precipitation and observed discharge. We control for snow by looking at snowfall records where 212
available; otherwise we only use events between April 1 and December 1. Spatial uniformity is 213
determined by using events in which precipitation between multiple rain gages is within 50%. Any events 214
with less than 10 mm of precipitation are excluded to minimize sensitivity to the initial abstraction 215
necessary to initiate runoff. The event discharge is calculated by computing the aggregate stream 216
discharge volume during the precipitation period and up to four days following the end of precipitation or 217
when event discharge returns back to pre-event baseflow levels. Because events are selected to be 218
temporally isolated, baseflow is accounted for by simply subtracting off the stream discharge on the day 219
immediately prior to the precipitation event (Qprior). 220
221
11
As has been observed previously, S is not a constant but varies with antecedent conditions. Prior work has 222
found that Qprior is a suitable estimator of antecedent conditions (Shaw and Walter 2009). Any 223
comparisons of S need to control for differences in Qprior. Thus, to relate changes in S to changes in land 224
cover, we can compare the S versus Qprior relationship among different eras. Because of scatter in this 225
relationship, we apply a boot strapping approach that constructs a linear, best-fit regression line for 5000 226
samples of the dataset for each era; we sample without replacement to generate the 5000 samples. From 227
this, we can establish the degree of uncertainty of the slope and intercept associated with each S versus 228
Qprior relationship for each era (Table 3). 229
230
Because there is not a direct, linear linkage between S versus Qprior and peak discharge (i.e. discharge is 231
determined from the non-linear Eqn. 1), we also calculated a peak discharge representative of a large 232
storm event (Qpeak) by randomly drawing a Qprior value, calculating an S by plugging Qprior into an S versus 233
Qprior linear regression relationship generated by boot strap, pairing this S value with a fixed precipitation 234
amount (51 mm), and calculating a Qpeak using the inverse of Equation 1 (i.e. Q=P2/[S+P] ). We used 235
uniformly distributed Qprior values between 2 and 4 mm d-1. The advantage of calculating Qpeak from back-236
calculated Qprior versus S values (instead of directly comparing the original Q values from the raw data) is 237
that we can control for differences in weather variables across eras. We generated 5000 Qpeak values 238
using the 5000 boot strapped S versus Qprior relationships for each era for each watershed. 239
240
3. Results 241
With the exception of the Sterling Creek, the watersheds exhibited sizable changes in land cover during 242
the period of analysis (Table 2, Figures 3 to 6). However, Sterling Creek had a rapid increase in forest 243
cover between the 1980’s (when stream gaging ended) and early 2000. Since we only analyze the period 244
when stream gaging records are available, we do not directly consider the 2000 land cover in Sterling 245
12
Creek in the analysis, but we do show it in Figure 5. As expected, this change in land cover across the 246
watersheds is dominated by a transition from agricultural to forested land cover. There was also an 247
increase in urbanized land cover across all the basins. This new “urbanized” land cover occurs in a 248
relatively rural setting and in most cases consists of single-family residential homes on large lots. As an 249
exception, Wappinger Creek, does have some concentrated commercial development in the southern 250
portion of the watershed near the cities of Poughkeepsie and Wappinger Falls. 251
252
We investigate the S versus Qprior relationship to assess the sensitivity of hydrologic response to changes 253
in land cover. In general, we would assume that a reduced sensitivity in hydrologic response would be 254
reflected by an increasing S value, corresponding to an increase in available watershed moisture storage. 255
Figure 7 displays the S versus Qprior plots for all four watersheds for each era in which they were 256
analyzed. Across all four watersheds, there are subtle visual distinctions among different eras. For 257
instance, in Wappinger Creek, above 0.8 mm d-1, the S values for the 1990’s appear to be systematically 258
larger than those of the 1930’s. Notably, in Limestone Creek the S versus Qprior plot appears to slightly 259
diverge from a linear curve and flatten at higher values of Qprior. We presume this may be due to the karst 260
geology providing some additional deep storage at high flows. However, because the linear regression 261
line fits relatively well (R2>0.64) and because there does not appear to be any change in outcome were the 262
non-linearity to be directly accounted for, we still only use the linear relationship to evaluate changes 263
among eras in Limestone Creek. 264
265
To more quantitatively assess possible differences between eras, one can examine the boot strapped mean 266
and standard deviations of the slope and intercepts associated with each S versus Qprior curve (Table 3). In 267
comparing intercept and slope means across eras within the same watershed, we assume that differences 268
in means greater than two standard deviations apart would indicate a significant statistical difference at 269
13
approximately the 5% level. In Fall Creek, differences in slope are well within one standard deviation of 270
each other, but intercepts between the 1930’s and 1990’s are nearly two standard deviations apart, 271
suggesting a meaningful shift in hydrologic response. A similar difference between slopes and intercepts 272
is also seen in Wappinger Creek between the 1930’s and 1990’s, suggesting a meaningful shift in S, and 273
thus hydrologic response, in this watershed too. Limestone Creek does not appear to have any significant 274
shift in slope or intercepts across eras. Sterling Creek has a moderate shift on the order of one standard 275
deviation in both slope and intercept. Due to the interaction of slope and intercept, it is ambiguous in 276
terms of which era has a higher S in the Sterling Creek watershed. Therefore, we also used the boot-277
strapped S versus Qprior relationship to directly estimate changes in discharge between eras for a fixed 278
sized precipitation event. This analysis can provide further insight into changes in hydrologic response in 279
Sterling Creek as well as further validate changes in the other three watersheds inferred by looking at 280
slope and intercepts alone. 281
282
In Table 4, we compare changes in discharge as based on the traditional tabulated Curve Number method 283
and as based on discharge values (referred to as Qpeak) determined from the boot-strapped S versus Qprior 284
relationships. For the calculations presented in Table 4, we evaluate the hydrologic response at a single 285
precipitation value of 51 mm, a moderately large precipitation event on the order of the one-year return 286
period storm in the locale of our watersheds. We replicated the calculations for precipitation values of 287
20.5 and 102 mm and found similar changes in hydrologic response, thus only the results for a 51 mm 288
event are shown. For Qpeak calculations, we indicate the fraction of the 5000 boot-strapped trials in which 289
the percent change in greater than zero. Fractions of Qpeak>0 near 50% indicate that the change is little 290
different from zero. Only when the Fraction of Qpeak>0 approaches zero or 100 does it suggest that the 291
change is statistically significant. 292
293
14
The traditional Curve Number method is meant to give some sense of typical expectations of how land 294
cover change would influence hydrologic response. In applying the traditional Curve Number method, we 295
use a Curve Number (CN) value of 55 for forest cover and 92 for urban land. Because of the range in 296
typical CN values related to agricultural and use, we assess two scenarios in which agricultural land is 297
assigned a CN of either 68 or 75. A CN range from 68 to 75 approximately spans CN’s related to pasture 298
and row crops for a class B hydrologic soil group. Based on the traditional Curve Number method, all 299
four watersheds see a decline in peak flow. Sterling Creek sees a limited decline because of the only 300
slight shift in land cover proportions. In all four watersheds, the traditional Curve Number method 301
suggests that declining agricultural land offsets any increases in urban land and that conversion to forest 302
land will result in at least some diminishment in peak flows. 303
304
The calculation of Qpeak results in very different outcomes relative to the Curve Number method. In Fall 305
Creek, Qpeak drops between the 1930’s and 1960’s, and then increases above the 1930’s value by 10.8% 306
by the 1990’s (Table 4). Greater than 90% of the trials have a positive increase, indicating a sizable 307
difference from a zero percent change. In Limestone Creek, Qpeak does not experience a change different 308
from zero between the 1930’s and 1960’s or the1930’s to 1990’s. In Sterling Creek, there is an average 309
decline of 22.8% in Qpeak with 84.5% of the trials having a negative decline. The large decrease in 310
Sterling is particularly surprising giving the limited change in land cover between eras, although as noted 311
earlier the land cover must be in transition given the large increase in forest cover by 2000. In Wappinger 312
Creek, there is a nearly 30% decline in mean Qpeak with 93.2% of the trials having a negative decline. The 313
large decrease in Wappinger Creek is in the same direction as might be expected given the traditional 314
Curve Number method, but of much greater magnitude. For the most part, the changes in Qpeak between 315
eras are consistent with changes in slope and intercept presented in Table 3. 316
317
15
4. Discussion 318
Based on the boot-strap based Qpeak, changes in hydrologic response differ among watersheds. More so, 319
these changes are not to the degree expected by the traditional Curve Number method and not necessarily 320
in the direction one would normally anticipate. We consider this boot-strap based Qpeak to be a particularly 321
robust means of assessing differences in hydrological response. Specifically, while there may be an 322
inclination to simply use a collection of direct discharge measurements for a given era (e.g. mean annual 323
peak flows during the 1960’s versus mean annual peak flows during the 1990’s), we suggest the approach 324
presented here is more robust because it accounts for variations in external, causative factors at the time 325
of discharge. In particular, it is quite evident from Figure 7 that antecedent moisture (as represented by 326
Qprior) strongly influences hydrologic response and a failure to account for this factor may obscure 327
perceived differences between discharge in different eras. The implementation of a computational 328
hydrologic model could be an alternative to our calculation and boot-strapping of S versus Qprior 329
relationships. However, the advantages of our approach relative to a more standard computational model 330
are that it is simple, relatively transparent, and it does not impose a model structure a priori (it instead 331
originates from the empirical relationship present in the data). 332
333
Of particular value in establishing the S versus Qprior relationships (Figure 7) is that the difficulty in 334
discerning differences in hydrologic response among eras becomes quite evident. While the basic 335
relationship between P, Q, and Qprior appears quite robust, there admittedly remains significant scatter. 336
Ideally, within eras, S values would display a near one-to-one correspondence to Qprior; instead for any 337
given Qprior value there is a nearly order of magnitude range of possible associated S values. This suggests 338
that there are other weather variables that remain unaccounted for. These variables likely remain difficult 339
to incorporate, especially in historic periods dating back to the 1940’s, and possibly include sub-daily 340
variations in precipitation intensity and fine-scale spatial variations in precipitation amounts. Thus, the 341
16
method provides insights into inherent limitations in discerning differences among eras and highlights the 342
difficulty in isolating land cover factors when weather variables vary among discharge events. 343
344
We cannot make a universal statement regarding the influence of land cover change on hydrologic 345
response given the varied changes in Qpeak among the four watersheds. But we can offer several specific 346
observations for why hydrologic response may have changed in certain ways in certain watersheds. In 347
Fall Creek, the moderate increase in Qpeak may be due to the moderate increases in urbanized land cover 348
(Table 2). Somewhat unexpectedly, the Limestone Creek watershed was found to have a larger change in 349
the degree of urban land cover between eras but only exhibited a marginal increase in hydrologic response 350
(Table 2). It is possible that features of urban land cover that cannot be distinguished from aerial 351
photographs account for this difference between watersheds. For example, the degree of connectivity 352
between impervious surfaces and stream channels may have evolved differently in the two watersheds 353
such as from differences in the extent of roadside ditch network and other drainage conveyances 354
(Buchanan et al. 2013). It is also possible that the presence of karst geology under the Limestone Creek 355
watershed provides enhanced storage relative to Fall Creek even as the watershed urbanizes. 356
357
Even with some urbanization, a doubling of forested land might still be expected to result in a change in 358
hydrologic response. But, the relative lack of sensitivity of hydrologic response to this change is not that 359
unexpected in glaciated watersheds in temperate, humid weathers. Most event discharge in these 360
watersheds is expected to occur due to saturation excess runoff generation (Walter et al. 2003, Dunne and 361
Black 1970). That is, runoff generation occurs on portions of the watershed where the full depth of the 362
pore space in the soil profile becomes filled. Runoff is typically generated near the base of hill slopes as 363
subsurface lateral flow from upslope areas combines with direct precipitation to saturate the soil column. 364
The degree of saturation excess runoff is primarily dependent on the depth to a confining layer and 365
17
porosity (Buda et al. 2009), factors which do not change significantly with land cover. Other work 366
supports the notion of the limited connection of hydrologic response to land use glaciated landscapes in 367
humid, temperate zones. Merz and Blӧschl (2009) found that runoff ratios in Austria were not well 368
predicted by curve number, and thus land use alone. Along these same lines, it has often been found that 369
the Curve Number works best when CN values are back-calculated and not taken from a table (e.g. Huang 370
et al. 2012). 371
372
We suggest that Wappinger Creek and Sterling Creek may have secondary factors besides land cover 373
which dominates changes in Qpeak. Forty-four small dams were constructed in the Wappinger Creek 374
watershed between 1950 and 1990. The exact amount of water that can be impounded behind these dams 375
for an extended period after storm events is not exactly known, but the NYS DEC dam inventory does 376
report the normal water level storage and the maximum water storage. The difference between total 377
normal storage and total maximum storage was almost 700,000 m3 (565 ac-ft). If spread over five days, 378
this storage could reduce peak flow by 0.30 mm d-1, partially accounting for the approximately 2 mm d-1 379
difference between Qpeak from 1930 to 1990. It is possible that the reported difference between normal 380
water level storage and maximum storage is only a rough estimate of available storage and that the dams 381
actually provide much greater storage that accounts for a larger portion of this 2 mm discrepancy. While 382
the other three basins did have several dams constructed in the time period during which hydrologic 383
studies were evaluated, the number of new dams was much smaller than found in the Wappinger Creek 384
watershed. Prior studies often note the disturbance in hydrologic regime due to large dams on main 385
channels (e.g. Fitzhugh and Vogel 2011, Poff et al. 2006), but the aggregate influence of small dams 386
(which individually impound stores of only several acre-feet) has not often been quantified (Vedachalam 387
and Riha 2013). 388
389
18
The behavior of Sterling Creek is less easily explained. Notably in this case, the negative change in Qpeak 390
is only present in 85.5% of all the bootstrapped trials (Table 4). Thus, there is some reasonable 391
possibility this change is within the range of uncertainty of the data. This watershed is also different from 392
the other three, having limited topographical relief, the likelihood of a high water table, and presumably 393
more active wetlands. Finer classification of land cover indicates that in the 1950’s 14% of the watershed 394
consisted of woody wetlands, primarily in riparian areas. It appears that by the 1980’s, much of this 395
woody wetland was identified as forest. It is possible this land was partially drained, potentially reducing 396
rapid runoff generation from these near stream areas. 397
398
These findings that reforestation (or presumably the reverse, deforestation) do not predictably influence 399
hydrologic response are in contrast to other studies, notably those in more tropical and sub-tropical 400
regions. However, such differences in outcomes are likely attributable to differences in hydrologic 401
processes in the different regions. In tropical and subtropical regions, forested watersheds do tend to have 402
very high surface infiltration rates and event flow tends to be dominated by lateral subsurface flow 403
(Elsenbeer 2001). However, when disturbed and converted to agricultural land, tropical soils often see 404
sharp declines in hydraulic conductivity. This has been attributed to intensive agricultural practices that 405
lead to soil compaction as well as a sharp decline in soil organic matter due to rapid rates of 406
decomposition and a failure to replace organic material (Recha et al. 2012). High soil organic matter 407
content is essential to maintain soil structure and high infiltration rates. In temperate regions, due to 408
cooler temperature, organic matter loss is much less and conversion to agricultural land (and back to 409
forest land) would likely see less change in soil structure. 410
411
5. Conclusions 412
19
Many existing studies of land cover change do not directly control for both variability in weather and 413
changes in non-land cover factors within watersheds. Additionally, many only examine dramatic changes 414
in land cover, not more moderate changes. To address limitations of existing studies, we longitudinally 415
assessed multi-decadal changes in four humid, temperate region watersheds using a variant of the runoff 416
ratio to account for weather variability. This is one of the few examples of such studies in the literature 417
that attempts to control for weather (by way of the runoff ratio variant) and innate watershed features (by 418
way of using a longitudinal approach). 419
420
A more refined understanding of the sensitivity of hydrologic response to land use is important for 421
informing decision making regarding land use management to reduce flood risks. Particularly with 422
weather change, there is an interest in devising adaptation strategies. While minimizing the extent of 423
urbanization may be useful, converting agricultural land to forest (the primary transition considered in 424
this study) was not observed to consistently reduce hydrologic response. This work suggests reforestation 425
alone may not always be an effective means to minimize future flooding in large glaciated, temperate 426
region watersheds. 427
428
Additionally, this work indicates that measures of spatially aggregated land cover may not translate into 429
meaningful measures of differences in hydrologic behavior. Quite simply, the runoff response in these 430
watersheds may not be strongly sensitive to land cover change, possibly in contrast to places where 431
infiltration excess overland flow is more dominant. Or, there may be important confounding features that 432
overwhelm the role of land cover. Such features may relate to connectivity of disturbed areas to the main 433
channel (Roy and Shuster 2009) or direct human influences on the hydrology such as withdrawals, 434
impoundments, or interbasin transfers (Weiskel et al. 2007). In this case, in one of our study watersheds 435
(Wappinger Creek), we found that a decrease in peak flow between the 1930’s and 1990’s corresponded 436
20
to the installation of numerous small dams in the same period. While hydrologists are certainly familiar 437
with the influence of large dams on major waterways, the aggregate impact of numerous small dams not 438
necessarily intended for flood control has not been extensively documented. There remains a need for 439
additional work to understand what type of fine scale features drive differences in hydrologic response 440
and to understand when land cover change will be superseded by changes in other landscape features. 441
442
Acknowledgments: Many thanks to Mark Storrings for providing technical assistance in georeferencing 443
aerial images. This work was supported by the New York State Water Resources Institute and the Hudson 444
River Estuary Foundation. 445
446
21
References 447
Arnold, J.G., J.R. Williams, A.D. Nicks, N.B. Sammons, 1990. In SWRRB—ABasin Scale Simulation 448 Model for Soil and Water ResourcesManagement. Texas A&M University Press, College Station, TX. 449
Baker, V.R., 1973. Geomorphology and hydrology of karst drainage basins and cave channel networks in 450 East Central New York, Water Resources Research, 9(3): 695-706. 451
Bathurst, J.C., A. Iroume, F. Cisneros, J. Fallas, R. Iturraspe, M.G. Novillo, A. Urciolo, B. de Bievre, 452 V.G. Borges, C. Coelle, P. Cisneros, J. Gayoso, M. Miranda, and M. Ramirez, 2011. Forest impacts on 453 floods due to extreme rainfall and snowmelt in four Latin American environments 1: Field data analysis. 454 Journal of Hydrology, 400: 281-291. doi: 10.1016/j.jhydrol.2010.11.044 455
Bates, C.G. and A.J. Henry, 1928. Forest and streamflow experiment at Wagon Wheel Gap, Colorado. 456 Monthly Weather Review, Supplement 30, USDA, Weather Bureau. 457
Beighley, R.E., and G.E. Moglen, 2002. Trend assessment in rainfall-runoff behavior in urbanizing 458 watersheds. Journal of Hydrologic Engineering 7(1): 27–34. 459 460 Boggs, J.L. and G. Sun, 2011. Urbanization alters watershed hydrology in the Piedmont of North 461 Carolina. Ecohydrology, 4(2): 256-264. 462 463 Bradshaw, C.J.A., N.S. Sodhi, K.S.-H. Peh, and B.W. Brook, 2007. Global evidence that deforestation 464 amplifies flood risk and severity in the developing world. Global Change Biology, 13: 2379-2395. doi: 465 10.1111/j.1365-2486.2007.01446.x 466 467 Buchanan, B., Z.M. Easton, R.L. Schneider, and M.T. Walter. 2013. Modeling the hydrologic effects of 468 roadside ditch networks on receiving waters. Journal of Hydrology, 486: 293-305. doi: 469 10.1016/j.jhydrol.2013.01.040 470 471 Buda, A.R., P.J.A. Kleinman, M.S. Srinivasan, R.B. Bryant, and G.W. Feyereisen, 2009. Factors 472 influencing surface runoff generation from two agricultural hillslopes in central Pennsylvania, 473 Hydrological Processes, 23(9): 1295-1312. 474 475 Burns, D., T. Vitvar, J. McDonnell, J. Hassett, J. Duncan, and C. Kendall. 2005, Effects of suburban 476 development on runoff generation in the Croton River Basin, New York, USA. Journal of Hydrology, 477 311: 266-281. 478 479 Dunne, T., and R.D. Black, 1970. An experimental investigation of runoff production in permeable soils. 480 Water Resources Research, 6(2): 478-490. 481 482 Elsenbeer, H., 2001. Hydrologic flowpaths in tropical rainforest soilscapes- A review. Hydrological 483 Processes, 15: 1751-1759. 484 485
22
Fitzhugh, T.W., and R.M. Vogel, 2011. The impact of dams on flood flows in the United States. River 486 Research and Applications, 27(10): 1192-1215. doi: 10.1002/rra.1417 487 488 Fry, J.A., M.J. Coan, C.G. Homer, D.K. Meyer, and J.D. Wickham, J.D., 2009. Completion of the 489 National Land Cover Database (NLCD) 1992–2001 Land Cover Change Retrofit product: U.S. 490 Geological Survey Open-File Report 2008–1379, 18 p. 491 492 Huang, M., J. Gallichand, C. Dong, Z. Wang, and M. Shao, 2013. Use of soil moisture data and curve 493 number method for estimating runoff in the Loess Plateau of China. Hydrological Processes, doi: 494 10.1002/hyp.6312 495 496 Lumia, R., 1991. Regionalization of flood discharges for rural, unregulated streams in New York, 497 excluding Long Island. USGS, WRIR 90-4197, Albany, NY. 498 499 Merz, R. and G. Blӧschl. 2009. A regional analysis of event runoff coefficients with respect to weather 500 and catchment characteristics in Austria. Water Resources Research,45, W01405. 501 doi:10.1029/2008WR007163 502
Poff, N.L.,B.P. Bledsoe, and C.O. Cuhaciyan, 2006. Hydrologic variation with land use across the 503 contiguousness United States: Geomorphic and ecological consequences for stream ecosystems. 504 Geomorphology, 79: 264-285. doi:10.1016/j.geomorph.2006.06.032 505 506 Recha, J.W., J. Lehmann, M.T. Walter, A. Pell, L. Verchot, and M. Johnson, 2012. Stream discharge in 507 tropical headwater catchments as a result of forest clearing and soil. Earth Interactions, 16: 1-18. doi: 508 10.1175/2012EI000439.1 509
Rose, S., and N.E. Peters, 2001. Effects of urbanization of streamflow in the Atlanta area (Georgia, 510 USA): A comparative hydrological approach. Hydrological Processes, 15: 1441-1457. 511 512 Roy, A.H. and W.D. Shuster, 2009. Assessing impervious surface connectivity and application for 513 watershed management. Journal of the American Water Resources Association, 45(1): 198-209. doi: 514 10.1111 ⁄ j.1752-1688.2008.00271.x. 515 516 Schoonover, J.E., B.G. Lockaby, B.S. and Helms, 2006. Impacts of Land Cover on Stream Hydrology in 517 the West Georgia Piedmont, USA. Journal of Environmental Quality, 35(6): 2123 – 2131. 518
Shaw, S.B. and M.T. Walter, 2009. Improving runoff risk estimates: Formulating runoff as a bivariate 519 process using the SCS curve number method. Water Resources Research, 45: W03404. 520
Sriwongsitanon, N. and W. Taesombat, 2011. Effects of land cover on runoff coefficient. Journal of 521 Hydrology, 410:226-238. 522
U.S. Soil Conservation Service (SCS) (1972), Hydrology, in National Engineering Handbook, U.S. Gov. 523 Print. Off., Washington, D. C. 524 525
23
Van Steenbergen, N. and P. Willems. 2013. Increasing river flood preparedness by real-time warning 526 based on wetness state conditions. Journal of Hydrology, 489:227-237. 527 528 Vedachalam, S. and S.J. Riha, 2013. Small is beautiful: State of the dams and management implications 529 for the future. River Research and Applications. doi: 10.1002/rra.2698 530 531 Vogelmann, J.E., S.M. Howard, L. Yang, C. R. Larson, B. K. Wylie, and J. N. Van Driel, 532 2001. Completion of the 1990's National Land Cover Data Set for the conterminous United States, 533 Photogrammetric Engineering and Remote Sensing, 67: 650-662. 534 535 Walter, M.T., V.K. Mehta, A.M. Marrone, J. Boll, P. Gerard-Marchant, T.S. Steenhuis, and M.F. Walter, 536 2003. Simple estimation of the prevalence of Hortonian flow in New York City watersheds, Journal of 537 Hydrologic Engineering, 8(4): 214-218. 538
Weiskel, P.R., R.M. Vogel, P.A. Steeves, P.J. Zarriello, L.A. DeSimone, and K.G. Ries III, 2007. Water 539 use regimes: Characterizing direct human interaction with hydrologic systems Water Resources Research, 540 43(4): W04402, doi:10.1029/2006WR005062. 541
Young, R.A., C.A. Onstad, D.D. Bosch, W.P. Anderson, 1994. AGNPS user’s guide. Agricultural 542 nonpoint source pollution model, version 4D03, July. USDA-NRS-NSL, Oxford, MS. 543 544
24
Tables 545
Table 1. Summary information for watersheds. 546
Watershed Total Area (km2)
USGS Gage No.
USGS Gage Period of Record
NWS Precipitation Gages
Fall Creek 323 04234000 1928 – present Locke, Cortland, Freeville Limestone Creek 219 04245000 1939 - 1986 DeRuyter, Morrisville Sterling Creek 114 04232100 1957 - 1995 Fulton, Oswego East Wappinger Creek 463 01372500 1928 - present Dutchess Ct. AP,
Millbrook, Wappinger Falls
547
Table 2. Percentage land cover type in each watershed in different eras. 548
1930’s 1950’s/1960’s 1980’s/1990’s
Fall Creek Forest 26.8 36.0 52.3 Agricultural 71.7 61.7 45.0 Urban 1.5 2.3 2.7
Limestone Creek
Forest 19.9 33.4 51.5 Agricultural 77.0 62.1 40.1 Urban 3.1 4.5 8.4
Sterling Creek
Forest --- 38.1 40.2 Agricultural --- 60.1 57.5 Urban --- 1.6 2.0 Open Water --- 0.1 0.3
Wappinger Creek
Forest 35.0 --- 65.0 Agricultural 64.0 --- 25.9 Urban 1.0 --- 9.1
549
550
551
552
553
554
555
556
557
558
25
Table 3. Mean and standard deviation of the slope and intercept of a linear regression line fit to 5000 559 trials of boot strapped S and Qprior . 560
Era
Slope Intercept Mean R2 Watershed Events
in Era Mean St. Dev. Mean St. Dev.
Fall Creek 1930’s 33 -1.100 0.075 4.815 0.103 0.85 1960’s 41 -1.120 0.063 4.899 0.089 0.83 1990’s 30 -1.167 0.109 4.572 0.112 0.77
Limestone Creek
1930’s 43 -1.618 0.157 5.753 0.103 0.69 1960’s 56 -1.484 0.149 5.567 0.128 0.65 1990’s 32 -1.444 0.190 5.500 0.149 0.64
Sterling Creek
1950’s 34 -1.113 0.106 5.100 0.211 0.72 1980’s 32 -1.007 0.105 5.370 0.162 0.74
Wappinger Creek
1930’s 49 -0.911 0.075 5.606 0.107 0.74 1990’s 33 -0.817 0.080 5.854 0.106 0.73
561
Table 4. Percentage change in watershed discharge between different eras when P=51 mm and Qprior is 562 between 2 and 4 mm day-1. The “CN Method” columns use land cover fraction reported in Table 2. Qpeak 563 is calculated from bootstrapped S vs. Qprior relationships. Fraction Trials % Change >0 indicates the 564 portion of the 5000 trials that result in a % change in Qpeak greater than zero. 565
Change in Q (when P=51 mm)
Watershed Transition
CN Method
(Ag CN=75)
CN Method
(Ag CN = 68)
Qpeak Fraction Trials % Change >0
Fall Creek 1930’s to 1960’s -4% -2% -3.7% 36.3 1930’s to 1990’s -13% -8% 10.6% 90.1
Limestone Creek
1930’s to 1960’s -9% -6% 2.1% 58.1 1930’s to 1990’s -16% -8% 3.3% 55.2
Sterling Creek 1950’s to 1980’s 3% -1% -22.8% 14.5
Wappinger Creek 1930’s to 1990’s -14% -6% -28.5% 6.8
566
26
Figure Captions 567
Figure 1. Location map of study watersheds and precipitation gages. 568
Figure 2. Representative comparison of aerial imagery and land cover classified as forest. This example 569 is taken from the Limestone Creek watershed due west of Cazenovia, NY during the 1930’s. The dark 570 band near the apex of the curve in State Rt. 20 is due to shadowing from a steep hillside to the west. The 571 dark band at the upper right hand corner is the border of an overlapping image. 572
Figure 3. Forest land cover in the Fall Creek watershed in the 1930’s is indicated in black. Forest land 573 cover added between the 1930’s and 1960’s is shown in dark gray while forest land cover added between 574 the 1960’s and 1990’s is the light tone. 575
Figure 4. Forest land cover in Wappinger Creek watershed in the 1930’s is indicated in black. Forest land 576 cover added between the 1930’s and 1990’s is shown in dark gray. 577
Figure 5. Forest land cover in the Sterling Creek watershed in the 1950’s is indicated in black. Forest 578 land cover added between the 1950’s and 1980’s is shown in dark gray while forest land cover added 579 between the 1980’s and 2000’s is the light tone. 580
Figure 6. Forest land cover in the Limestone Creek watershed in the 1930’s is indicated in black. Forest 581 land cover added between the 1930’s and 1960’s is shown in dark gray while forest land cover added 582 between the 1960’s and 1990’s is the light tone. 583
Figure 7. Storage versus Qprior curves for the four watersheds. Different symbols indicate different eras. 584
585
27
586
Figure 1. Location of study watersheds and meteorological stations within New York State. 587
588
28
589
Figure 2. Representative comparison of aerial imagery and land cover classified as forest. This example 590 is taken from the Limestone Creek watershed due west of Cazenovia, NY during the 1930’s. The dark 591 band near the apex of the curve in State Rt. 20 is due to shadowing from a steep hillside to the west. The 592 dark band at the northeast corner is the border of an overlapping image. 593
594
595
29
596
Figure 3. Forest land cover in Fall Creek in the 1930’s is indicated in black. Forest land cover added 597 between the 1930’s and 1960’s is shown in dark gray while forest land cover added between the 1960’s 598 and 1990’s is the light tone. 599
30
600
601
602 Figure 4. Forest land cover in Wappinger Creek watershed in the 1930’s is indicated in black. Forest land 603 cover added between the 1930’s and 1990’s is shown in dark gray. 604
31
605
Figure 5. Forest land cover in the Sterling Creek watershed in the 1950’s is indicated in black. Forest 606 land cover added between the 1950’s and 1980’s is shown in dark gray while forest land cover added 607 between the 1980’s and 2000’s is the light tone. 608
32
609
610 Figure 6. Forest land cover in the Limestone Creek watershed in the 1930’s is indicated in black. Forest 611 land cover added between the 1930’s and 1960’s is shown in dark gray while forest land cover added 612 between the 1960’s and 1990’s is the light tone. 613
33
614
Figure 7. Storage versus Qprior curves for the four watersheds. Different symbols indicate different eras. 615
616
617
618