You are on page 1of 23
Rows, 2aaioe WCT TIO THA) 2+ — ‘ARRAYS ‘tte = | euston ona ot ror thebyer cet yer G ram concen mm ren nymutcennevy YEON bores sere a Fig. 2:9. Schematic diagram of «full hrdimanstnal mel (aap om Medal and Facnugh, 1969), All hydrstagaphic ute are represented by che or nan ayes 3.3 Laying Out the Grid In a numerical model, the continuous problem domain is replaced by a dise cretized domain consisting of an array of nodes anc associated finite difforance blocks (cells) or finite elements. The nodal grid forms the framework of the numerical modal. The cancoptual model and the selection of model type es discussed in Sections $.1 and 3.2 will determine the overall dimensions of the grid. Tho selection of olthor a finito difference or firite element code influences the structure of the grid ‘TYPES OF GRIDS There are two types of finite difference grids: the biock-centerod grid (Pig. 43.108) and the mesh-contezod rid (Fig, 3.10c). The difference between them lies mainly in the way in which flux boundaries e7e handled, In the block- dis. the oes ‘the Fig. hem ock- centered approach flux boundaries always are located at tae edge of the block. In 2 mesh-centered grid, the boundary coincides with a sede. Details on the ‘way in which boundarios are reprosented in a finite differonce model are pre sented in Chapter 4 1n lange general computer codes, the finite differerce mathematios for boundaries are more easily treated with the block-centersd appsoach. Conse- ‘quontly most codes, including MODELOW, use this type o” grid, PLASM uses a block-centered approach that allows the user to switch fre black-centered 0 mash-contered trostmont of boundaries (Chepter 4, Box 42). : é ay gf STREMEHEY -~"| RRDUNO Rope, (a) rcblem domain. Nod boumdenee exe designated atthe swackin range ww along the Soeurlnae, Tho rivor lully penerass the equer and may be copieelod by v onstat bead ound [o) Boci-contered tata dren pd, No ow boundene are losated the def backs, The Specint-ead vor bousdaey Is loeted on nace The gr lege cath prblem dancin, Ie) Nest-ewntrad fate difeeaces. Both no-ou ane enosantshea brandarios fl cls the en {a} Tongulr finite elements Node cumbersome soa: lement mumbo oe ezled Rot mo ow and apecia-hond tomndatos all dvely on the aes, (0) Caadnlatal Hots slomonts Neds number ee shown elem nrbar ave cde Bh no low and speiiod head Laurdan fll Sry on tke andes. ea 6 ; Hees Pweeet *Daep]* | raw (G10 g, SRRBIER, | Fig, 3.10 (continued) Finite elements allow more flexibility in designing @ grid. Two-dimen. sional elements are either tangles or quadrilataals (Figs. 8.10d,0 and 3.11). limensional elements are tetrahedvons, hexahedrons, or prisms (Fig ‘he nature of the interpolation (basis) function used to define heacs ‘within the element determines whether the elenent is linear, quadratic, or cubic. The most commonly usod oloment is the incor olomont. AQUIFEM- uses linear triangular elements exclusively. Some codes (e.g, FREESURP, by ‘Neuman, 1976) ellow a mix of trianglos and quadtilaterals, DEFINING MODEL LAYERS ‘The soloction of modol type will detsrmino whethor one or moro than one model layer is coquired. Kf only one layer is needed, this layer typically repre- sons a single hydrostratigraphile unit. Tho concopiual model is usod to halp decide how mang layers are needed. ‘Whon quasi threc-dimensional models are used to represent rosional flow systems, hydrogeologie units are assumed to be horizontal or, in other words, hhave zero dip. Most geologic units slope at some angle to the horizontal (Fig. 3.13a)-In many cases the slope (dip) is quite small (1 or 2 dagroes) and tho units ccan be represented by horizontal model layers (Fig. 3.13b) for modeling pus- AAA ouaprame cuse b UNEAR ——quapamic cunie acer © picuaoranic sicueIs ig, 2.14. rwosdiaaceioual ota samen (Hajakors and Pides, 912) (@) Teanguler elements (0) Quaillatcnslements (Seren dipy tay) (6) Qulaterlchmonts (grange em) poses, Horizontal modal layers that represent dipping uns are shovn in Fig. $.14, Pinching out of aquifers or confining beds in qussi Uhcee-dimensional simulations can be handied as illustrated in Fig. 9.15, where changes in lithol- ogy within the layor are accommodated by changing tie transmissivity of the layer or the Ieakance of the confining bed. It necesmry, profile and full three-dimensional models may be used to simulate dipping units as shown in Fig. 9.16, po gs Fie 21% (b) Hohedons te) Ps me een, Taveg] 7 wwconenen | REIRESENTER NoREeTLY sama Rarer IZ = LAYER? ‘CONFINED ‘UNCONFNED (AGuFER 2 Fig. 3.12, Dotition of modo! Ivars in ctoss secon. (Ge) Dipping goclogie units are cent! er hycroewctmerhe ut (6) Reprecention of te dipping unit in (se lpene n qual thy» aaeslonal mode. The ‘nls ee tpedaod os Losuostl yarn ee thet in the model the eos eye a liga wit the principe dlectens ofthe ydrauie conductivity toc. Agar te represented ae ayer 1 fans unconinad. Agua 2s repressed ss consinn, seceptwhera X ope oat be coining bea isnot modeled expliely bot ie reprenented by = lakags tem that ba function of the Lad Aiffeeance acs Une confining bed nnd the vertical hydrals conduct and thickness of the ‘confining bed. ORIENTING THE GRID ‘present, should be alluned with K, (Fig. 3.17), However, itis not always practical to do this. In Fig. 9.19, for exemple, model layers are horizontal to the surface although the bedding is at an angle to the surface; hence, the 2 axls is not parallel to K,. Likewise the vertical axes forthe grids shown in Pig. 3.16 aro not perpendicular to bedding planes. In such cases It is typically assumed that the angle between the dip of the beds anc the hosizontal as is small so that Ke cam be assumed to be nearly colinear with the vertical axis If itis not possible to aliga the grid with the principal directions of the hydraulic conductivity tensor end if the anisotropy effects are determined to be Important, the govern- ing equation should be written to include the off-diagonal componcnts of the hydraulic condnetivity tensor (Pig. 2.3). Some finite olsment codes allow for these condi Alene ae uae Laver 2 - Upper Potemec aauiier VD TA] constant head ESSSI No tiow EEEEED Leacance vetweon confining Fig. 3:44 Recreselation of dipping hydostuoirapic wil in Deleware using = quss are fisvenstons mere Pl ps 1897, Gxcundwater Reve dbwenrard fom tho unetafed Calon isan: (ayer te the tee unite tho Potomac agus. Lakage of wats rim dhe uncoating fer scouted hom erlaaad spon’ haus Lounda pode nlaer {and ean ales Uiigee! lo somes tho turcnieton Ghote o€ natal Iornn Inyern 1 end 2. The ‘onet f tho hse e lavas wth th wean syste ssluletd by the specie eds und Irakanes provartios assigned tthe a shaded ectanguter sts hese les 2nd ayer 2 Those tres havo eda boundaries to connect to the layers by lecknce ove aesociated wih tho fourth ide —— n-ne ptt Ty Ty neater COE nfs nies a 1 Ques fe Sereno ecient ‘rows shaw yrical ard horzona!earnection Witt edeu engi ay rows teorocens b Present —S Avsent——__$ a 5 ¢ [ Ff td aan ret reine ay se [4 case 2Aauier Pnchout init Casita Moco! Conesctataton Ly HH { Fig. 25. Represent of pichouts (ea 1982). lal Plschaut ef» confining beds sulted in 9 cuas thren-imeeniondl model The conan odie raprvontea by lower eskonoe terme whan tke eal ba i sen! (N-C), Whee no oufzng sata are prsat (A-U), feakanceferas betweealaye ue aealte! bese the ‘amas mous of ale hydraulic eondact vite nd tickers 8x 2, (by Pinchow fan agua sila i fl thre smenaonel model, Cantining unis andthe sullen ate roprseatd by modelers. Tas plnesow ofthe etfs preted by changing Invdrogeoiogie properties a layer 2 Tare! th change tom ano to confining bed rot that fall outside the boundaries of the modeled are Tiag andor a Clo lnc‘ whoras Codes et all hi modeled area are active nodes. Inactive nodes are not part of the solution but 4, sill uso up storage space tn the arraya nooded by the cade. The problem of ° inactive nodes arises because finite difference pis are resangalar while the ee ares to be modeled frequently isnot [oy Fig. 2.178. a Finite element grids do not have inaclive nodes because the elements are oe fitted exactly to tho Boundary og, Fig. 2170). Whon itis necoasary to simulate athe interaction between the groundwater aystem end the boundaries, it way be dant Caitcal to approximate the boundaries as closely ab posssle. An exemple of dlaver such a simulation is ono involving the calculetion of flow into (or out of) a ance . S008 |F @ Sand ore revel ® Sandstone © Limestone Z ee a oe : (gonumze>) aci-e ts j Fig. 3.17. (continued) surface weter body (Fig, 3180,b). megular boundaries car: be simulated with special finite difference approximations (Remson ot al, 1971; Rushton end Redshaw, 1975) but those be taken fat the node falls diroctly on the boundary when using a of the blocks (Huyakorn and Pinder, 1985, pp. 346-348) ‘When concern focuses on the interior of the grid, the boundaries may be set fr from the area of interest so that imposed stresses lo the interior part of the system do not each the boundaries. Ia such a slmulation, fiting the grid to the exact shape of the boundary is not crucial to the modeling =ffort For example, Fig. 3.194 shows a nite difference grid used to simulate the effects of pumpin from municipal water supply wells in Madison, Wisconsin, The wells pump from a sendstone aquifer that is part of a lange regional aquifer cystom extond- ing inta eastem Wisconsin, Ilinois, and ows. Therefore, when boundaries are placed sufficiently far from the conter ofthe grid, the effects of pumping do not reach the boundaries within the time period usec in the tansient simulation. DISTANCE FEET) DISTANCE (FEET) Fig. 3.18 amples of sna nodal specingim fate element gs nese the bounday wrth ureee trae bodle. (Gl Fe elemene gd foe a peninaua in Wisconsin, const with « mtx of tanger and ‘quailatera hermes (rad 192) Fie mal pacing se orem the shal aren Ariae gonadal sheng to Gren Bey (3) Ageid showing the ure of ll model spacing sar abacar wall dasgae ls pote! econ fenctorocat in thwesera lay lms hah watt ble wen tho Pa River Dede. Ta Po River foensthe roster boundery end ho siver bane alo ssecontod by ane id Te novihos abd ‘eae boundaces ru specie head and the estem exdwestrabounderic aa a0 Bow bond ‘os (Gbolal, Teleo, and Ulla, Water Saou seach, 27) 3p W0S€13, 1986 cxpy gM bythe Amaican Geophysical Uren tae lar and smclost River oun e038 Dy Fig 3.48 (continsed) SPATIAL SCALES "ner nodal spacing will Similany, the change in head in ofthe vrtloal nodal spacta, ' jowever, if there are significant vertical head gredients, two or more lyors should bo used to roprocent single hydrostratigraphfc unit. Variation of aquifer properties in the horizontal dimension will usually accu over a longer length interval than variations in the vertical dimension. Finally, tho variability in areal recharge, pumping, and recharge or discharge to rivers should be con- iN ig. 2.18. xample of iroqulr fast france gids (6) The pd fore model of Dune County, Waconte (MEL 1973). Foe grid apecing inthe {Rte of the gy ead ocepesené mull gating, Noow boordarien are located fa columns re 88 ad ts ws 1 ed 50, hich ee no% shoe i the ire fh) Agidangued io ecommcda' slay spaced nodes yer proetion wells end sults Sows {no tafaton orn the Sarahece River te & slow equer fo outeern Now York Stale Fag 06) Detail of the grid used io the ental porno tka mada cca ae shove in the [c)'A gid designed io provide Ear nodal epcing nara wharo pumping esse are expected 0 cou forthe la eae ia acrheatrn Aeon Bee aad ychaner, 988) AM alle ae 2fow ‘ate ete SS, Fig, 3.18. (contimed sidcrod. Smaller grid spacing will be needed to represent river nodes and pumping nodes. It is conceivable thet areal recharge may change significantly from ono cell or clement to anothor, but typically vary little information is, ayailable on field-measured cecharge rates and a constan: value is usually as- ‘sured over large portions of the gtd. ‘The overall sizeof the mociled aroa will also affect tne selection of nodal spacing. A arid with a small number of nodes is preferred in order to minimize date handling and computer storage and computation tire. Yet, itis desirable to use a largo numbor of nodes to cepresont the systom accurately. The nced to select meaningful boundaries may require modeling a lerge area. A compro- tise between accuracy and practicality is necessary. One way to resolve the trade-off between number of nodes and required level of dcteil is to use a technique sometimes called telescopic mesh refinement (Ward st al, 1987) ‘whoreby 6 coarse grid is used to model a large problem domain bounded by the physical lunits of the aquifer system, The solution is used to define subregional BQ overoermarncas 21m _mpreer Fig. 2.90 (commued) xf ze 2B 8 ' Eby 3 i ean. Doha t Poet zone i 21 ree i qi Pp ly R Bais cae PEltilla:.: 8 z BoikG 2338 6 Sanigiig ithe abs Piei bliss atiLdtieg: ghidyeae fiR ae Ess}: + Fig. 318 [contimod) a ee ee oundaziee, which then define a new smaller provlem domain. The telesooping ies until the gcid is smell enough to oblainthe desired dial (Section 4.2 and Box 4.3) Finite Difference Grids Nodes are labeled using an (i,j,k) indexing convention (Fig, 3.9), to refer to the position within a row, colurm, and layer. A twe-dimensional finite difference ri is gonoratod by specifying arrays of values for Ax and Ay for the horizontal Plane, in a three-dimensional simulation an aay of Az valuos may also be Specified, althoush MODFLOW uses a different approach for discretizing the ‘vertical dimension (Box 3.1). ‘Nodes may be regulerly spaced so thet Ax, dy, and A are all constants but not necessarily equol to each other. More commenly, howover, an iregular grid ja necessary in order to use small grid spacing in one sree of the problem Homain, Tho gtid ia designed s0 that nodes are closely spaced where steep hydreulic gradionts are expected (¢ 8. noar pum2ing Wells), or to define narrow Heets or other hydrologiealy sigalficant features. Exemples of Site difference tide vith iregularly spaced nodes aro shown in Fig. 3.19, ‘When stoall grid spacing is used in the interior of the grid, it may be ocossiry to increase the nodal apecing as the grid is expanded out to the Bitar Fs ae fea model nlf amo miei Te roe al pacing ne mare hen 13 Gries the previous nod For examplo if the smallest nodal spece is ane meter, the next space should be ‘po mors than 1.5 meters. The next space beyane It should be no mare than 2.25 | meters, and so on. A factor of two may be used ‘ar e fow rows and columns but | fis advisable not to expand the entire grid usig a fsctor of two. T ‘gid expansion arfsas because the finite difforence expression Bialvalive have larger extar when darived for regular grid spacing, The Anite difforenco oxpressi gular grid is ecrrot to the fst onder, fora finite diffcronce oxpressio: is correct to the second om | Thed in truncation error con be domonstrated theoretically by means | ofa Taylor sees cxponsion (Romeon etal: 1672) a houriatic explanation of the cause of tho largor error when dealing with jnreguler nodal spacing follows. Consider tie one-dimensional system shown. in Fig. 2.20, For an irtegular grid Ax is not @ constant and the second decivative In the aeighborhood of (is the diffarence inthe fist derivatives as calculated at pointe 11/2 andi + 1/2 (Pig. 3.20), Using the notation defined in Fig. 3.20, the second derivative may be approximated «follows: (20) deh) a (ab) |) eh z] 2h ak 2b) ining naz othe fonts! sabe the shut raid ‘bler sr003 is a finite ans swith hhown ative ated 3.20, (32a) (3.26) Fea eve Fig, 9.20 Finite difecoce ids whore 1 + 2 rptesents the osation halfway bobrasa i and ii 1 stheloeton halfway beorers|~ Land Ati is the didtacce between nodes and 1+ 4:8 is tho distance Botoon nodes i ~ 1 and {and xe tho loath of hel (a) Regular red for hic Sts = Se = Ata (8) termini for whlch ra # dq # So Note that flsst derivatives are calculated at j + 1/2 and i~ 1/2, When Ax is a ‘constant (Fig, 1.204). the polnta {+ 1/2 end I~ 1/2 coincide with the edges of the nodal block so that the nodal point (i) is centered bween i ~ 4/2 and 1 + 1/2, Howover, for tho irvogular grid shown in Fig. 3.20b, the location of, | + 1/2 does not coincide with the edge of the nodal block that contsins node i and nodal point (i) is no longor contered botwoon t ~ 1/2 ani i+ 1/2, Tho finite difference solution calculates the head st the point halfway between i — 1/2 and i + 1/2, When the node is not contorad betwoen i ~ U2 and i + 1/2, tho location for which the head is calculated does nol coincide with the location of the nodo, thereby introducing somo arror into the solution. Note that for illus ‘ration purposes, the grid shown in Pig. 3.20b was expanded using a factor of {our rathor than the recommended factor of 1.8. te Element Grids A disadvantage of finite element models is that the input of date required ta define the grid ts more laborious than for finite difference madels. Finite ele- ‘mont models require that each node and element be numbered (Fig. 3.1040) 1! that the coordinate location (x, y, 2) of each node and the node numbers WY re arid tnd the last nonzero entry, inclusive, along any row of the matrix. The band- ‘width is oqual to 2 = (SBW — 1). The semibandwidth mey be calculated from sBw=R+1 G3) ‘whore Fis the maximum difference in eny two node numbers that define a single element within the grid (Istok, 1989). For example, n the geld shown In Fig. 3.21c, Bis 17, the seribandwidth is 18, and the barcdwict t dimens Jase to unity. This requirement is si factor of 1.5 used in expanding fnito difforonco grids and is necessezy to mini- mize numerical e:xors, For example, numerics) errors can be :oinimized by exclusive use of equilateral uiangular elomerts. Experience has shown that 1 sapect ratios grea ‘should be oyoided. Furthcrmor a trensition z j ‘caBlan FU GONE RIUG ela elamssue lay areal TPL OA, dealing with anisctropic matorals, the shape clthe-elemen's should be consid~ ered in t] jivalent transformed isotropic damain and designed so that the tapos ate tn tho cope dawn dvs nat excnd fer Fer dts on enw forming from an anisotropic to an isouopic domain see Freeze aad Chesry (1979, pp. 174-178). Because design ofa finite clement grid can be quite time Consuiming, the use of « preprocessor to assist in geic generation anc optioal node numboring is highly Tocommended for complex prablome (og. Te clare, 1989). 3.4 Assigning Parameter Values DATA NEEDS Data needed for groundwater flow models are summacized in Table 3.1. The data can be grouped into two goneral categor'es. Data under category A. the physical framework, define the geometry of thesystem including the thickness. and areal extent of cach hydrostratigraphic unit. Hydrogoologic data includo information on heads and fluxes (Items B.1 and B.2 in Table 3.1), which are needed to formulate the conceptual model end check model calibration (Chep- | tor), Hydrogeologic data also define aquiler properties and hydrologic stresses (items B.4-B.5 in Tablo 3.1). In addition, the distribution of effective porosity is, required when calculating average linear velocities from head data for input 16 ‘tacking codes (Chapter 11). Obiaining the information necessary for malin Is not an easy task. Some data may be obtained irom oxisting reporis, butin most cases additional on-site field work will be required. A discussion of th fleld techniques for acquiring these data is beyond the scope of this book. Only a briof overview of method ologies wil be presented below. Transmissluity and storaye coefficient azo typlcally obtained from pumping test rosults (Walton, 1987). For modeling at a local scale, values of hydraulic conductivity can be determined by pumping tosts if volume-averaged valuos are dosired or ay slug tests if point values are required (Bouwer, 1989; Bouver ancl Rice, 1998; Hverslev, 1951). For uncon solidated sand-size sediment, iydraulic conductivity may also be obtained from laboratory grai size analyses (Masch and Jenny, 1966) or from laboratory pormoability tosts using pormeamotors, Pormcenotor results must be used vrith caution because hydraulic conductivity values oblained from permeameter tosts typically ore sovoral ordors af magnitude smaller then values measured in situ (eg, Tanaka and Hollowell, 1969; Herzog andl Morse, 1904; White, 1988), The discropancy is caused by roarcangoment 2f grains during repacking the semple info the permeameter, Fucthermore, large-scale features such as {race tures, gravel lonsos, or other types of bedding that may impart transmiasion.

You might also like