TABLE 5.

Seasonal E. coli (target variable) regression tree split criteria^a for terminal nodes^b

Data set (total no. of samples)c	Root node split criterion	Secondary split criteriond	Median E. coli density (CFU 100 ml−1) ± MAD	No. of samples
E. coli densities for spring (595)	NUD_CATTLE BARN, >2.29 km	DPAST_2K, >1.26 obs. km−2	58 ± 1,040	190
	NUD_CATTLE BARN, >2.29 km	DPAST_2K, ≤1.26 obs. km−2	20 ± 206	315
	NUD_CATTLE BARN, ≤2.29 km	SHREVE, >5.0	108 ± 258	61
	NUD_CATTLE BARN, ≤2.29 km	SHREVE, ≤5.0	230 ± 516	29
E. coli densities for summer (750)	SHREVE, >15.5	NUD_CATTLE BARN, ≤3.65 km	112 ± 114	87
	SHREVE, >15.5	NUD_CATTLE BARN, >3.65 km	20 ± 88	383
	SHREVE, ≤15.5	NUD_PAST, ≤1.16 km	468 ± 3,650	141
	SHREVE, ≤15.5	NUD_PAST, >1.16 km	174 ± 521	139
E. coli densities for fall (575)	FORAGEP_2K, >0.39 km2 km−2	NA	230 ± 608	64
	FORAGEP_2K, ≤0.39 km2 km−2	NUD_FORAGE, >2.95 km	24 ± 84	73
	FORAGEP_2K, ≤0.39 km2 km−2	NUD_FORAGE, ≤2.95 km	78 ± 526	438

^aRoot node split criterion, the variable and condition by which all the data were divided into two nodal groupings (child nodes); secondary split criterion, the variable and condition by which the child nodes derived from the root nodal split were divided (for purposes of brevity, we present only results up to this tree level in this study). The variables that define the E. coli split criteria are described in Table 1.

^bTerminal nodes are data groupings where no further splitting occurs.

^cAll data sets are for 2004 to 2007. Note that winter data are not included, since CART could not cross-validate that seasonal data set, as a result of lack of data structure.

^dobs., observations; NA, not applicable, since there was no terminal node of data at that level in the tree model.