Environ. Sci. Technol. 2005, 39, 7226-7233
Models for Hydrologic Design of Evapotranspiration Landfill Covers VICTOR L. HAUSER,* DIANNA M. GIMON,† JAMES V. BONTA,‡ TERRY A. HOWELL,§ ROBERT W. MALONE,| AND JIMMY R. WILLIAMS⊥ Victor L. Hauser, P.E., 13214 Hunters Breeze, San Antonio, Texas 78230, Mitretek Systems, 16414 San Pedro Avenue, Suite 340, San Antonio, Texas 78232, North Appalachian Experimental Watershed, Agricultural Research Service, U.S. Department of Agriculture, P.O. Box 488, Coshocton, Ohio 43812, Conservation and Production Research Laboratory, Agricultural Research Service, U.S. Department of Agriculture, P.O. Drawer 10, Bushland, Texas 79012, National Soil Tilth Laboratory, Agricultural Research Service, U.S. Department of Agriculture, 2150 Pammel Drive, Ames, Iowa 50011, and Texas Agricultural Experiment Station, 808 East Blackland Road, Temple, Texas 76502
The technology used in landfill covers is changing, and an alternative cover called the evapotranspiration (ET) landfill cover is coming into use. Important design requirements are prescribed by Federal rules and regulations for conventional landfill covers but not for ET landfill covers. There is no accepted hydrologic model for ET landfill cover design. This paper describes ET cover requirements and design issues, and assesses the accuracy of the EPIC and HELP hydrologic models when used for hydrologic design of ET covers. We tested the models against highquality field measurements available from lysimeters maintained by the Agricultural Research Service of the U.S. Department of Agriculture at Coshocton, Ohio, and Bushland, Texas. The HELP model produced substantial errors in estimating hydrologic variables. The EPIC model estimated ET and deep percolation with errors less than 7% and 5%, respectively, and accurately matched extreme events with an error of less than 2% of precipitation. The EPIC model is suitable for use in hydrologic design of ET landfill covers.
Introduction The technology used in the landfill cover portion of landfill remediation is changing, and an alternative coversthe evapotranspiration (ET) landfill coversis coming into use (1). Conventional landfill covers (Figure 1) use barriers to control water movement downward through the cover. Important design requirements for conventional landfill covers are prescribed by Federal rules and regulations (2), and a model is accepted for their design (3, 4). However, * Corresponding author phone: (210) 493-7527; fax: (210) 4790482; e-mail:
[email protected]. Mitretek Systems, retired. † Mitretek Systems. ‡ North Appalachian Experimental Watershed. § Conservation and Production Research Laboratory. | National Soil Tilth Laboratory. ⊥ Texas Agricultural Experiment Station. 7226
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 18, 2005
FIGURE 1. Comparison of conventional, barrier-type and ET landfill cover cross-sections. design requirements for alternative landfill covers are not prescribed by rules and regulations. Furthermore, there is no accepted model for ET landfill cover design. This paper describes cover requirements and design issues and assesses two engineering models for use in ET landfill cover design. Requirements of ET Landfill Covers. The concepts of the ET landfill cover were first described in 1994 (5, 6). The ET landfill cover consists of a layer of soil covered by native grasses or other native vegetation; it contains no barrier or “impermeable” layers (Figure 1). The ET cover uses two natural processes to control infiltration: the soil provides a water reservoir, and natural evaporation from the soil plus plant transpiration (ET) consumes the water stored in the soil water reservoir. The ET cover is a practical, easily maintained biological system that will remain effective during extended periods of timesperhaps centuries. It offers opportunities for improved performance and low construction and maintenance cost. The proof of the ET landfill cover concept and important design requirements were described by Hauser et al. (7). ET landfill covers generally must meet the following requirements for landfill covers: (a) control infiltration of precipitation into the waste, (b) isolate the wastes from receptors and control waste movement by wind or water, and (c) control landfill gas. A cover meeting the first requirement also isolates the waste for most sites. Gas produced within the landfill may be controlled by conventional methods, if needed. The engineering design model should focus on hydrologic design to assess the control of percolation into the waste. The ET cover has the following minimum requirements: (a) support a robust, healthy vegetative cover, (b) support rapid and prolific root growth in all parts of the soil cover, and (c) hold enough water in the soil to control water movement below the cover during extreme or critical events (7). Because of the design requirements, the hydrologic design of ET landfill covers requires analysis of complex interactions between climate, soil, and plants (7, 8). A comprehensive computer model is needed to evaluate an ET cover’s daily hydrologic response; it should evaluate soil, plant, and climate variables, including their interactions and influence on daily water balance. The ET cover is expected to last for decades, possibly centuries; therefore, the model should estimate longterm performance which may require generating daily climate variables. The design model should require only parameters that can be measured or are available in historical records for the site. It should not require calibration because sitespecific data needed for adequate calibration are seldom, if ever, available. Hydrologic Water Balance. The first requirement of a landfill cover is to control the amount of precipitation that percolates through the cover and may enter the waste; it is called deep percolation (PRK). Because it is part of a complete system, one must estimate the entire hydrologic water balance to determine PRK. 10.1021/es048020e CCC: $30.25
2005 American Chemical Society Published on Web 08/05/2005
TABLE 1. Basis for ET Estimates by EPIC and HELP
FIGURE 2. Water balance terms for an ET landfill cover (eq 1). The hydrologic water balance for an ET landfill cover may be expressed as (Figure 2):
P ) ET + Q + L + SWS + PRK + error
(1)
in which P ) precipitation (includes irrigation, if applied); ET ) actual evapotranspiration; Q ) surface runoff; L ) lateral flow; SWS ) change in soil water storage; PRK ) deep percolation; and error ) lack of balance in the equation. L is usually small and is assumed to be negligible. Where all terms are measured (e.g., lysimeter measurements), the difference or lack of balance is an expression of measurement error. A major focus of model evaluation for ET covers is the accuracy of daily estimates of PRK. However, the estimate of PRK is strongly affected by errors in estimates of daily values of the other water-balance terms ET, Q, and SWS. The water-balance estimate is controlled by the model’s use of site-specific parameters, including those for climate, soils, and plants (7, 8). Site-specific climatic factors that are important include daily measurements of precipitation, maximum and minimum temperature, relative humidity, solar radiation, and wind. The SWS is a large part of daily water balance and errors in model estimates of SWS influence estimates of ET, Q, and PRK. The potential value of daily evapotranspiration (PET) (recently called reference evapotranspiration) is the basis for ET estimates (9, 10). The actual ET for the site may be estimated from the PET value. The accuracy of PET and ET estimates is the biggest controlling factor for hydrologic modeling accuracy because ET is the largest water consumption factor on the right side of eq 1. Soil stores precipitation within the ET cover, it is the medium in which plants grow, and it provides nutrients for plant growth. Soil-water storage capacity is controlled primarily by soil physical properties, including the soil bulk density and texture. The soil-water storage volume available on any day is controlled by the cumulative balance between infiltration and water removal from the soil by ET and percolation. Purpose. The purpose of this work was to evaluate fully developed and tested models for use in engineering design of ET landfill covers. The models should be capable of estimating the major input and output terms of the water balance (P, ET, Q, and PRK). Accurate evaluation of a model for design use requires field-measured data for P, ET, Q, and PRK and should include an assessment of the accuracy of the field measurements. Others evaluated models (11-15). None of these evaluations met all of the requirements stated in the purpose above for field-measured data or models. They did not use independently measured values of ET. We tested two engineering models that estimate all factors important to ET landfill cover design. We evaluated the HELP model, version 3.07 (3, 4), and the EPIC model, version 8120 (16). While these models have different origins, both of them
characteristic
EPIC
HELP
estimates actual root growth soil density vs root growth soil temperature vs root growth
yes yes yes
no no no
evaluate the hydrologic cycle and satisfy basic requirements for engineering design. EPIC Model. The development of the environmental policy integrated climate (EPIC) model began in the early 1980s (16-18). EPIC estimates the hydrologic water balance, including Q, PET, ET, SWS, and PRK. It uses a daily time step to simulate climate and hydrologic parameters for a wide range of soils, climates, and plants. EPIC uses readily available input data and can simulate hydrologic response for hundreds of years. The EPIC model was extensively tested for water balance estimates in dry and wet climates, including sites with significant accumulation of snow in winter (17). Testing of the EPIC model against measured field data demonstrated that it estimated PRK with satisfactory accuracy (19-21). HELP Model. The hydrologic evaluation of landfill performance (HELP) model (3, 4) is widely used and accepted for design of conventional, barrier-type landfill covers with bottom liners. The primary purpose of the HELP model is to provide water-balance data with which to compare design alternatives. The HELP model uses climate, soil, and design data to estimate daily landfill hydrologic performance as expressed by surface storage, snowmelt, runoff, infiltration, ET, soil moisture storage, leachate recirculation, and leakage through barrier layers. It is capable of modeling landfill systems for up to 100 years. The HELP model was designed to evaluate barrier-type covers, but it has not met expectations for the evaluation of vegetative covers (22, 23). Model Differences. There are significant differences between the models. The estimate of ET dominates hydrologic modeling accuracy because it is the largest term in the water balance and it controls the size of the other terms estimated by the model. Major factors that control the model estimate of ET for both models are listed in Table 1. Because the mass of plant roots in a soil layer limits how much water plants can remove from the layer during each day, root mass and rate of growth are important. Root mass and growth rate processes are controlled by soil density and temperature. High soil density reduces soil-water holding capacity. Additional information regarding ET estimates may be found in refs 7-10 and in the Supporting Information for this paper. Another significant difference between the models is that the HELP model assumes that frozen soils are impermeable; however, the EPIC model assumes that they have reduced permeability. This has the potential to affect runoff estimates.
Test Data and Methods We evaluated the EPIC and HELP models using accurate field measurements that were available from the Agricultural Research Service (ARS) of the U.S. Department of Agriculture at two locations. At Coshocton, OH, the ARS measured the hydrologic response of meadow for a total of 17 years with a lysimeter. At Bushland, TX, ARS measured the hydrologic response of alfalfa and corn for two years in a dry climate with two lysimeters. The soil profiles in each lysimeter at both sites are hydrologically similar to ET landfill covers. We utilized lysimeter and climate measurements for each day of each calendar year during the 17 years at Coshocton and for two sets of biennial periods at Bushland. The evaluations were based on measured, daily values of weather VOL. 39, NO. 18, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7227
parameters for the sites except for one data set. There was an apparent error in the solar radiation measurements at Coshocton, OH during the 1970-1979 period. Because the models can estimate daily solar radiation from other climate input data, we set both models to estimate daily solar radiation during 1970-1979 (8). Weighing and recording lysimeters are capable of measuring all terms of the water balance directly (P, ET, Q, and PRK). The high-quality lysimeters and equipment used at each site are described in the Supporting Information and in refs 24-32. They weigh the mass of the lysimeter and both ET and P may be calculated as the difference between successive measurements. Percolation from the soil and surface runoff were continuously measured. At both sites, all parts of the water balance were measured hourly; however, we used daily summaries of the data to test the models. The input data are described in the Supporting Information. Coshocton Data. The Ohio measurements were from lysimeter Y101d, located at the North Appalachian Experimental Watershed (NAEW), ARS, USDA, about 16 km (10 mi) northeast of Coshocton, OH, at 40.4° N lat and 81.5° W long. The elevation of the lysimeter surface is about 361 m (1185 ft) above sea level and about 5% of the annual precipitation occurs as snow (24). The average annual precipitation is 950 mm (37 in.) at Coshocton and the soil may remain frozen and snow-covered for several weeks during the winter. The vegetation was meadow and similar to plant cover that might be established on an ET landfill cover in that region. The measurements were made with a weighing and recording monolithic lysimeter. The dimensions of the soil block contained in the lysimeter are 4.3 m (14 ft) long, 1.9 m (6.2 ft) wide, and 2.4 m (8 ft) deep, with the long dimension up and down hill. The lysimeter soil block is an undisturbed natural soil profile from the site; it includes bedrock in the bottom layers, thus ensuring natural percolation processes. The lysimeter mimics hydrologic conditions of the surrounding watershed. The land slope is about 23% and the lysimeter precision was 0.25 mm/day (24-26). Precipitation, air temperature, humidity, wind, and solar radiation measurements were available from a nearby weather station and precipitation was measured at the site. Percolation outflow was about 31% of precipitation. Bushland Data. The Texas measurements were from two lysimeters located at the Conservation and Production Research Laboratory, ARS, USDA, Bushland, TX. The lysimeters are located on the Texas High Plains, in a semi-arid climate, about 24 km (15 mi) west of Amarillo, TX, at 35.2° N lat, and 102.0° W long. The elevation of the site is 1170 m (3840 ft) above sea level. The average annual precipitation is 455 mm (18 in.) at Bushland; the site is windy, and has low relative humidity and high PET. At the Bushland site, the climate is dry to semi-arid, and snow is a small part of the water balance; winters are cold enough to kill or create dormancy for most plants during several months. The two weighing and recording, monolithic lysimeters (27) contain undisturbed columns of Pullman clay loam soil with surface area of 9 m2. The soil depth was 2.3 m. Each lysimeter was installed in an irrigated field. The cropped area surrounding the irrigated fields was similar for at least 1 km in the north-south direction and about 0.5 km eastwest. The predominant wind direction is south or southwest. Thus, there were no edge effects in the water-balance terms measured by the lysimeters. The surface slope at the site and over the adjacent fields is typically less than 0.5% and the lysimeters allow no surface runoff. The lysimeter precision was 0.045 mm/day. Precipitation, air temperature, humidity, wind, and solar radiation measurements were available from a weather station operated at the site (over irrigated grass) and also 7228
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 18, 2005
TABLE 2. Total Water Balance Errors for Measurements by the Lysimeters location
total total total input output errora (mm) (mm) (%)
daily error (mm)
precipitation measured by lysimeter Coshocton, meadow, 1970-79 11067 11456 3.5 0.11 Coshocton, meadow, 1987-93 7170 7226 0.8 0.02 Bushland, alfalfa (2 years) 2953 3013 2.0 0.08 Bushland, corn (2 seasons)b 1664 1662 -0.1 >-0.01 precipitation measured by rain gauge at the lysimeter Coshocton, meadow, 1970-79 9949 11456 15.1 0.41 Coshocton, meadow, 1987-93 6487 7226 11.4 0.29 Bushland, alfalfa (2 years) 2875 3013 4.8 0.19 Bushland, corn (2 seasons)b 1568 1662 6.0 0.13 single rain gauge reading 0.25 a Error as a percentage of (precipitation + irrigation). b Water balance error for May 1 to December 31.
from another station at laboratory headquarters (over mowed, native grass). Percolation outflows were small or zero. The soil at the site and in the lysimeters is the highly productive Pullman clay loam found on several million hectares of the southern high plains of Texas, New Mexico, and Oklahoma (28). An important feature of Pullman soil is the natural high-density subsoil layer from 0.45 to 1.8 m depth; the bulk density is equal to or greater than 1.6 Mg m-3. The high density of the subsoil layer restricts root growth and water extraction (29). Irrigated corn grew in one lysimeter during 1989 and 1990, and irrigated alfalfa grew in the other lysimeter during 1996 and 1997. Howell et al. (30) described a corn experiment conducted at the site during 1987 and provided details regarding lysimeter operation and crop management. The 1996 alfalfa season was the first growing season after its fall seeding in 1995. Irrigation amount was measured by lysimeter catch and confirmed by measurements of the volume of water applied. Both lysimeters were heavily irrigated; however, the opportunity for deep percolation resulted in only small amounts of deep percolation under corn and none under alfalfa. Little or no percolation was expected because the clay soil at the site has excellent water holding properties and irrigation was applied by sprinklers. Lysimeter Measurement Errors. The lysimeters at Coshocton and Bushland, along with their associated infrastructure, are among the best such facilities in the world. However, field measurements used to evaluate models should first be evaluated for their accuracy because errors in the data limit the accuracy of the evaluation. The accuracy of the data used in these evaluations may be assessed in two ways: (i) accuracy can be based on the precision of a single lysimeter measurement; (ii) the cumulative error of the measurements can be estimated using eq 1. Daily measurements of each term of the water-balance equation, except for the error term, are available for each lysimeter. We calculated daily values of the error term (eq 1). Table 2 contains the total water input and output for the period of each record, along with the error of the totals and the average daily error derived from the cumulative measurements. The lysimeters more accurately measure snowfall and rainfall than the nearby rain gauge (31, 32). The lysimeter at Coshocton caught about 10% more precipitation than a nearby rain gauge (Table 2). The difference was between 2.5% and 5.8% at Bushland. The range of error for measurements using lysimeterestimated precipitation is -0.1% to +3.5% (Table 2). When using rain gauge data in the balance, the error ranges from 5 to 15%. The difference between water-balance errors of
rain gauge data and lysimeter data is smaller at Bushland than at Coshocton, probably because of the difference in snowfall between the sites. The average daily errors (Table 2) were generally similar to or less than the precision of a single rain-gauge reading suggesting small cumulative errors in the measurements. Where the water balance was estimated from lysimetermeasured precipitation (Table 2), the average daily error was near or less than the respective lysimeter precision, indicating that the lysimeter measurements were of good quality. Rain-gauge data are the only precipitation data normally available for ET cover design. However, model test results can be no better than the data used in testing, so we used the higher-quality lysimeter precipitation measurements for this model evaluation to provide the most accurate comparison between models. Model Tests. Design engineers will seldom, if ever, have the data needed to calibrate a model. Therefore, we used the models as published and entered the measured and available data for each site (8). We obtained HELP version 3.07, from the U.S. Army Corps of Engineers website. We used nine sublayers within layer type one (top layer) to simulate the soil of the lysimeter. The HELP model uses the modified Penman equation for PET estimates, and the curve number method to estimate surface runoff. We obtained EPIC version 8120 (EPIC8120) from the Texas A&M University-Blackland Research Center web site. The EPIC model used 10 soil layers to describe the soils of each site. The EPIC model offers four methods for estimating PET; we chose the Penman-Monteith method because it is accurate and applicable to both humid and arid sites (9). We chose the curve-number method to estimate surface runoff.
Model Evaluation The emphasis of this evaluation is on deep percolation (PRK). Actual ET and Q together are much larger than PRK, and errors in model estimates of either affect the accuracy of PRK estimates. Therefore model accuracy in estimating ET and Q may define a model’s usefulness. The measured values and estimates by the models of parameters in eq 1 were daily values. However, because they are compatible with statistical evaluation, we evaluated annual sums of ET, Q, and PRK for each year. The extreme or critical hydrologic event is important because it could cause deep percolation through the cover. The extreme event may last for 1 day or for several days. We evaluated the annual maximum of the monthly sums of PRK within a year as a surrogate for the extreme design event. Many preferred statistical measures are based on the assumption that the data come from a normally distributed population, so we assessed the normality of each measured parameter (8, 33). We selected summary and comparison statistics to evaluate model performance for both normally and nonnormally distributed parameters. Summary Statistics. We used the mean and percent error to describe normally distributed data and the median and percent error to evaluate nonnormal parameters. We defined the percent “error”, as follows:
error )
[
]
P m - Om × 100 reference
(2)
in which Pm ) predicted (modeled) mean or median, Om ) observed (measured) mean or median. We defined the reference value in two ways: as the measured parameter value or as the measured lysimeter precipitation plus irrigation. Percent error is an important measure of model performance; however, the reference value determines the size of
the “error” term and influences the interpretation of results. For example, the error of the PRK estimate by HELP for Bushland-Corn is -16.5 mm/year. The percentage error based on the measured PRK value is -75%, while the error based on total precipitation is -2.0%. Although the obvious, intuitive assumption is that measured values should be used in estimating error, there are valid reasons for using total precipitation as the reference value in error analyses. There are three important issues involved. (i) Small parts of the hydrologic water balance, such as PRK, are measured directly and independently in lysimeter measurements. However, model estimates of PRK are not independent and contain increased error as a result of errors made by the model in estimating the larger terms. (ii) The relative size of waterbalance terms is important. Even though the error of PRK, for example, may be only a few millimeters, the percent error may be large when calculated from a small measured amount. (iii) It is important to define the error in a way that is consistent with the intended use of the model estimates and the data available to the designer. A major concern in landfill cover design is the fraction of annual precipitation that may infiltrate through the cover and into the waste; thus, annual precipitation is a logical reference value. We used both measured parameter values and precipitation as references for error estimates. Comparison Statistics. Comparison statistics evaluate how well the model estimates represent the corresponding measurement. We assessed the normality of each parameter at the 10% significance level (33) to provide guidance in using statistics. Analysis of measurements revealed that the annual totals and the maximum month totals for each year for ET and PRK were normally distributed; but the Q measurements were not normally distributed. Comparison statistics selected for evaluating the normally distributed estimates were the following:
RMSE )
x
∑(P - O )
100 O h
n
2
i
i
i)1
(3)
n
n
n
∑(O - Oh ) - ∑(P - O ) ) 2
( EF )
2
i
i
i)1
i
i)1
(4)
n
∑(O - Oh )
2
i
i)1
n
∑
( CRM )
n
Oi -
i)1
∑P ) i
i)1
(5)
n
∑O
i
i)1
in which RMSE ) normalized root-mean-square error %; EF ) modeling efficiency; CRM ) coefficient of residual mass; Oi and Pi are the observed (measured) and predicted values of each evaluation datum i; n is the number of observed and predicted value pairs; and O h is the mean of the observed values (20, 34, 35). VOL. 39, NO. 18, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
7229
TABLE 3. Criteria Chosen to Assess Model Performance statistic
optimum
satisfactory result
error (%) RMSE (%) EF CRM
normal data 0.0 0.0 1.0 0.0
-20 e error e 20 0.3 -0.2 < CRM < 0.2
error (%) MdAE (%) REF
nonnormal data 0.0 0.0 1.0
-20 e error e 20 0.3
FIGURE 3. Average, monthly surface runoff at Coshocton, 19701979.
Comparison statistics selected to evaluate nonnormally distributed estimates were the following:
MdAE ) median(|Oi - Pi|: i ) 1, 2, ..., n) ×
[
( ) 100 Om
(median(|Oi - Om|: i ) 1, 2, ..., n) median(|Oi - Pi|: i ) 1, 2, ..., n)) REF ) median(|Oi - Om|: i ) 1, 2, ..., n)
]
values (Table 4); however, the EPIC model produced much more accurate estimates than the HELP model. When evaluated against precipitation, the errors in Q were less than 6% (calculated from Table 4); that error is less than the waterbalance errors of the field measurements. Three of the four estimates for PRK by the EPIC model met the performance criteria, while all of the HELP estimates failed the evaluation (Table 4). When evaluated against precipitation, the errors in PRK estimates by EPIC met the performance criteria; some errors for the HELP model estimates were up to 15% (calculated from Table 4). The model errors were consistent with the lysimeter errors (Table 2). Additional insight into model estimates of Q may be gained from an examination of total runoff measured at Coshocton over the ten-year period from 1970 to 1979. The HELP model predicted 10 times and EPIC predicted 5 times the measured runoff for the 10-year period. Figure 3 shows average measured monthly runoff amounts and estimates by EPIC and HELP for the Coshocton lysimeter during 1970-1979. Both EPIC and HELP use the SCS curve number (CN) method to estimate surface runoff. The data in Figure 3 suggest that the CN chosen for use in the models was appropriate during the warm months of the year. The models estimated more runoff than was measured during winter and springsthe time when snowmelt and frozen soil influence surface runoff. The HELP model treats frozen soil as impermeable; however, the EPIC model treats frozen soil as though it has reduced permeability, so it produced a better estimate of runoff (Figure 3). Annual, Maximum Monthly PRK. Extreme values of PRK are important when evaluating ET landfill covers because they define critical design requirements. Where the reference
(6)
(7)
in which MdAE ) normalized median absolute error, %; REF ) robust modeling efficiency; and Om ) the median of the observed values (20 and 34). We chose assessment criteria for each summary or comparison statistic to aid the evaluation of model performance (Table 3). The criteria are somewhat arbitrary; however, they provide uniform guidelines to assess model performance and they are similar to published values (20).
Results The models were not calibrated. They were run with the available data, duplicating their use in an engineering design. Annual Summaries. The ET estimates by the EPIC model satisfied the performance criteria when error was based on the measured parameter; however, the estimates by the HELP model did not meet the criteria for the Coshocton data (Table 4). All ET estimates satisfied the performance criteria for percent error based on annual precipitation (calculated from Table 4); however, the errors for the HELP model were as large as 20%, the maximum value for acceptance. Both models produced large percentage errors for Q estimates at Coshocton when evaluated against measured
TABLE 4. Summary Statistics for Mean Annual Totals of Model Estimated ET, Q, and PRKa measured
b
P
ET
Q
PRK
model
mean (mm/yr)
mean (mm/yr)
median (mm/yr)
mean (mm/yr)
modeled ET
EPIC HELP
1107 1107
767 767
4.4 4.4
EPIC HELP
1024 1024
764 764
1.2 1.2
EPIC HELP
1476 1476
1514 1514
0b 0
0 0
Bushland, alfalfa 1460 1478
-4 -2
0 0
EPIC HELP
832 832
809 809
0 0
22 22
Bushland, corn 867 869
+7 +7
0 0
mean (mm/yr)
modeled Q
modeled PRK
median (mm/yr)
error (%)
Coshocton, 1970-1979 368 753 -2 368 547 -29
3.4 71.0
Coshocton, 1987-1993 276 732 -4 276 570 -25
0.0 7.2
error
mean (mm/yr)
error (%)
-23 +1500
318 492
-14 +34
-100 +500
259 429
-6 +55
0b 0
0 71
0 >999
0 0
31.0 5.5
+41 -75
a Percent error of model estimates based on measured parameter value. Italicized values exceed the criterion for satisfactory results (Table 3). Bushland lysimeters prevented runoff and the models predicted none.
7230
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 18, 2005
TABLE 5. Summary Statistics for Mean of Annual, Maximum Values of Monthly PRKa mean, annual maximum monthly PRK error measured as:
Pb
diff.c
precip.d (%)
measd.e (%)
model
ann. (mm)
EPIC HELP
1107 1107
Coshocton, 1970-1979 102 122 20 102 205 103
2 9
20 101
EPIC HELP
1024 1024
Coshocton, 1987-1993 79 88 9 79 145 66
1 6
11 84
EPIC HELP
1476 1476
Bushland, alfalfa 0 0 0 0 47 47
0 3
0 >999
EPIC HELP
832 832
Bushland, corn 24 2 4 -18
