Predicting the extent of lakeshore development using GIS datasets

Figures & data

Figure 1 Location of lakes >20 ha in Minnesota. Docks were counted on 150 highlighted lakes, 50 in each of 3 ECS provinces that contain numerous lakes.

Figure 2 A section of shoreline showing the automated-counted dock polygons and points along with Minnesota Land Use and Cover (MNLU) developed raster cells. The 2 circles represent a 30 m and 75 m search radius around the black shoreline point in their center. This point would be classified as developed according to both the automated-counted dock data (dock within 30 m) and the MNLU data (developed cell center point within 75 m). The background shows the 2008 National Agricultural Imagery Program (NAIP) aerial photo.

Table 1 Measures of classification accuracy derived from confusion or error matrices, composed of the count data in the first 4 rows (Metz 1978, Fielding and Bell 1997).

Measure	Calculation
Hits: True Positive (TP)	Number correctly classified positive
False Alarms: False Positive (FP)	Number classified positive that are actually negative
Correct Rejections:True Negative (TN)	Number correctly classified negative
Misses: False Negative (FN)	Number classified negative that are actually positive
Positive Predictive Value (PPV; also called precision)	TP/(TP + FP)
Negative Predictive Value (NPV)	TN/(TN + FN)
True Positive Rate (TPR; also called sensitivity)	TP/(TP + FN)
True Negative Rate (TNR; also called specificity)	TN/(TN + FP)
Accuracy	(TP + TN)/N, where N = sum of all cases

Table 2 Selected characteristics for 150 Minnesota lakes as a whole and separated by ecoregion: Laurentian mixed forest (Forest); eastern broadleaf forest (Transition) and prairie parkland (Prairie). Developed shore points/km uses generated points along the shoreline classified by calculating their proximity to the nearest development indicator for each dataset: manual-counted docks, automated-counted docks, Minnesota Land Use and Cover (MNLU), National Land Cover Database (NLCD), and a composite dataset containing developed cells from each of the 2 raster datasets (Combined).

				Median developed shore points/km
	Median Area (ha)	Median Shoreline (km)	Median Manual- Counted Docks/km	Manual- Counted Docks	Automated- Counted Docks	MNLU	NLCD	Combined
All	76.3	4.9	1.3	2.5	1.4	4.2	7.3	10.1
Forest	70.0	5.5	1.4	2.8	2.1	3.9	4.7	6.1
Transition	55.4	4.6	3.9	6.1	3.5	11.0	11.4	17.6
Prairie	102	5.4	0.5	1.0	0.7	2.9	6.5	8.6

Figure 3 Total operating characteristic (TOC) curves using statewide data from a semiautomated image analysis of docks (Docks), Minnesota Land Use and Cover (MNLU), National Land Cover Database (NLCD), and Combined MNLU and NLCD datasets. The total number of hits+misses is shown by the horizontal line at the top of the plot; the gray diagonal line shows the expected curve for a random relationship; and the gray triangles are mathematically impossible areas for the TOC curve. The vertical line shows the point where the number of points classified as developed equals the actual number of developed points. Solid dots on each curve mark the 75 m cutoff point, and the open square marks the 30 m cutoff point for Docks.

Figure 4 Total operating characteristic (TOC) curves using statewide data from a semiautomated image analysis of docks (Docks), Minnesota Land Use and Cover (MNLU), National Land Cover Database (NLCD), plotted separately by ecoregion: (a) Laurentian mixed forest (Forest); (b) Eastern broadleaf forest (Transition); and (c) Prairie parkland (Prairie). In each plot, the number of hits+misses is shown by the horizontal line at the top of the plot; the gray diagonal line shows the expected curve for a random relationship; and the gray triangles are mathematically impossible areas for the TOC curve. The vertical line shows the point where the number of points classified as developed equals the actual number of developed points. Solid dots on each curve mark the 75 m cutoff point, and the open square marks the 30 m cutoff point for Docks.

Figure 5 Scatter plots of developed shoreline point density for 150 lakes, classified using (a) automated-counted dock data (Docks); (b) Minnesota Land Use and Cover data (MNLU); and (c) National Land Cover Database (NLCD); and. The dashed lines show a 1:1 relationship, and the solid lines are linear regression lines.

Table 3 Testing shoreline classification using 3 prioritization scenarios to evaluate amount and distribution of undeveloped shoreline. Lakes were selected using the automated-counted dock layer (Auto-Dock), the Minnesota Land Use and Cover (MNLU) data, and the National Land Cover database (NLCD) according to the criteria listed and compared to lakes selected using the same criteria derived from the manual-counted dock data, which represents the true development state.

	(a) Goal: Identify pristine lakes for protection (Criteria: >90% undeveloped shoreline)
Dataset	Lakes Identified	TP	FP	TN	FN	PPV	NPV	Accuracy	TPR	TNR
Auto-Dock	93	86	7	56	1	0.92	0.98	0.95	0.99	0.89
MNLU	67	60	7	56	27	0.90	0.67	0.77	0.69	0.89
NLCD	46	41	5	58	46	0.89	0.56	0.66	0.47	0.92
(b) Goal: Identify large segments of undeveloped shoreline on highly developed lakes for protection (Criteria: >40% of shoreline developed AND maximum undeveloped length >800 m)
	Lakes Identified	TP	FP	TN	FN	PPV	NPV	Accuracy	TPR	TNR
Auto-Dock	8	6	2	134	8	0.75	0.94	0.93	0.43	0.99
MNLU	21	11	10	126	3	0.52	0.98	0.91	0.79	0.93
NLCD	23	3	20	116	11	0.13	0.91	0.79	0.21	0.85
	(c) Goal: Identify highly fragmented lakes for shoreline restoration projects (Criteria: average undeveloped length <300 m)
	Lakes Identified	TP	FP	TN	FN	PPV	NPV	Accuracy	TPR	TNR
Auto-Dock	32	32	0	107	11	1.00	0.91	0.93	0.74	1.00
MNLU	20	18	2	105	25	0.90	0.81	0.82	0.42	0.98
NLCD	12	11	1	106	32	0.92	0.77	0.78	0.26	0.99

*Abbreviations in table headings: TP = True Positive, FP = False Positive, TN = True Negative, FN = False Negative, PPV = Positive Predictive Value, NPV = Negative Predictive Value, TPR = True Positive Rate, TNR = True Negative Rate

Metz CE. 1978. Basic principles of ROC analysis. Semin Nucl Med. 8:283–298.

 PubMed Web of Science ®Google Scholar