A new logis
tic regression (LR) model was used
to predic
tthe probabili
ty of ni
tra
te con
tamina
tion exceeding 4mg/L in predominan
tly shallow, recen
tly rechargedgroundwa
ters of
the Uni
ted S
ta
tes. The new model con
tainsvariables represen
ting (1) N fer
tilizer loading (
p &l
t; 0.001),(2) percen
t cropland-pas
ture (
p &l
t; 0.001), (3) na
tural log ofhuman popula
tion densi
ty (
p &l
t; 0.001), (4) percen
t well-drained soils (
p &l
t; 0.001), (5) dep
th
to
the seasonally highwa
ter
table (
p &l
t; 0.001), and (6) presence or absence ofunconsolida
ted sand and gravel aquifers (
p = 0.002). Observedand average predic
ted probabili
ties associa
ted wi
thdeciles of risk are well correla
ted (
r2 = 0.875), indica
ting
tha
t the LR model fi
ts
the da
ta well. The likelihood of ni
tra
tecon
tamina
tion is grea
ter in areas wi
th high N loadingand well-drained surficial soils over unconsolida
ted sandand gravels. The LR model correc
tly predic
ted
the s
ta
tus ofni
tra
te con
tamina
tion in 75% of wells in a valida
tion da
tase
t. Considering all wells used in bo
th calibra
tion andvalida
tion, observed median ni
tra
te concen
tra
tion increasedfrom 0.24
to 8.30 mg/L as
the mapped probabili
ty ofni
tra
te exceeding 4 mg/L increased from
![](/images/en<font color=)
ti
ties/le.gif">0.17
to &g
t;0.83.