Sacramento River Vegetation Map

Summary: 
This project will be used to post reports and other materials associated with DFG/Calfed Ecosystem Restoration Program Grant ERP062002.

This project will be used to post reports and other materials associated with DFG/Calfed Ecosystem Restoration Program Grant ERP062002.

Project Contact: 

Subtask 2.1.1 - Map Validation

The Sacramento River vegetation map was validated over the summer (June-September) of 2008. Field validation of CSUC polygons was conducted along the 100 mile stretch between Red Bluff (south of the diversion dam) and Colusa on over 30 properties owned by the Department of Fish and Game, The Nature Conservancy, Fish and Wildlife Service, and the Department of Water Resources. Polygon accuracy was checked visually by editing polygon attributes in ArcPad 7.1 loaded into Trimble GeoXM units, via rapid assessments, intensive modified Whittaker plots, re-digitizing 500m by 500m blocks of vegetation, input from other project collaborators, and by revisiting and referencing Vaghti (Vaghti 2003) relevé plots. Data analysis was conducted in JMP IN 7.

Rapid Assessments

Rapid assessments were collected along 100 miles of the Sacramento River between Red Bluff and Colusa. We used the CNPS protocol as outlined in the Vegetation Rapid Assessment Protocol created by the CNPS Vegetation Committee (http://cnps.org/cnps/vegetation/). RA’s were then used to validate the CSU Chico map and will be turned into the DFG vegetation program.

Intensive Sampling

Intensive modified Whittaker plots were set up throughout the project area, a 100 mile stretch of the Sacramento River from Red Bluff to Colusa. These plots were set up in areas that were floristically representative of a particular vegetation type at the association level. Data collection was conducted based on procedures developed by Barnett and Stohlgren (2003).

Visual Field Validation

Field validation of CSUC polygons was conducted along a 100 mile stretch of the Sacramento River between Red Bluff (south of the diversion dam) and Colusa on over 30 properties owned by the Department of Fish and Game, The Nature Conservancy, Fish and Wildlife Service, and the Department of Water Resources. Polygon accuracy was checked visually in the field by species dominance and then editing polygon attributes in ArcPad 7.1 loaded into Trimble GeoXM units.

Digital Map Check

Using ArcGIS9.2, we created 500m x 500m blocks that cover the study area. Blocks overlapping > 33.4% of the 2007 Riparian Map were selected at random for re-digitization. GIS Technicians, after a period of training and cross-calibration, digitized vegetative cover with each selected block using the 2007 aerial photos as reference for interpretation. In all, X blocks were re-digitized for (see map above) for statistical analysis. Statistical analyses consisted of paired comparisons, first for inter-rater reliability, and second for vegetation class representation.

Subtask 2.1.2 - Cross-walk Comparison and Calibration

We assisted CSU Chico with methodological development of Subtask 2.1.2, which will allow a comparison between the two mapping efforts (1997 – 2007). The intent of this task was to create an iterative quality assurance (QA) procedure to examine the consistency of map class types between two time periods for the purposes of change detection. The initial methodology consisted of generating 103-5 random point locations (x, y coordinates) within the universe of mapped vegetation, which were then attributed with class labels from the two data sources and inspected for frequency of association. The tabular cross-walk was to be amended accordingly until an appropriate frequency threshold has been achieved (e.g., 90% stasis by class between two time periods). Upon completion of this task, we determined that it was necessary to create 1.4 x 106 stratified random points to achieve any statistical consistency. This cross-walk table is available for download below in the technical report.

Subtask 2.2.1 - Change Detection

To measure trends in riparian composition between the two time periods and two datasets with confidence, four (4) data matrices will be constructed and analyzed. Each polygon will be negatively buffered (i.e., area reduced via perimeter reduction) to account for 0m, 1m, 5m, and 10m boundary confidence intervals and will provide an objective measure of the degree (% change) of observable change within riparian composition independent of cartographic effort. In other words, if 15% of a polygon is still coded differently for each of the four levels of confidence (0m ... 10m), there is no detectable cartographic boundary influence. On the other hand, if there is 20% change at 0m and 2% change at 1m, there is considerable boundary effect and comparative statistics should only be generated at negative buffers greater than 1m (e.g., 5m or 10m).

Subtask 2.3.4 - Dataset analyzing vegetation species composition, relative canopy cover and frequency

Floodplain structural and compositional vegetation data were collected to determine the structural biodiversity of floodplains along the Sacramento River. Thirteen floodplain transects were collected from August to September 2008 along the Sacramento River between Red Bluff and Colusa. Each transect traversed multiple land-age classes, substrates, and vegetation types. We collected structural data including tree occurrence information as well as species compositional data via intensive modified-Whittaker plots. A GPS point was taken at each tree larger than 10 cm dbh within 2 meters from the transect line. Tree species, tree height, diameter at breast height (DBH), and crown radius of each tree were entered into TerraSync using Trimble ProXT. Species composition was analyzed through intensive modified Whittaker plots located in each unique vegetation type or floodplain age along the transect line.

Outreach: ESRI Proceedings Paper

Abstract

We quantified the map accuracy for the Sacramento River Monitoring and Assessment Project to help land and water manager’s better plan for restoration efforts. While map errors are quantifiable and even predictable, linking the causes of error to complex environmental and geographic variables would improve decision making. We evaluated patterns of GIS-induced map error on over 32,000 acres based on environmental and GIS variables like floodplain age and edge complexity. We conducted extensive field validation and used spatial statistics to compare environmental variables with vegetation map inaccuracies. We then constructed a multivariate model to predict errors in certain vegetation types. We field validated 15% of map polygons (n=8,067) which were 85% correct (K=0.83). Using validated polygons, we found errors occurred most frequently on older floodplains but rates varied by vegetation type. By incorporating error in attribution and spatial assignment, restoration planners have a more realistic assessment of current conditions.

Citation

Viers, JH, AK Fremier, and RA Hutchinson. 2010. Predicting map error by modeling the Sacramento River floodplain. Proceedings from the 2010 ESRI International User Conference, San Diego, California. 21 ppd.