2003 Progress Report: Integrating Economic and Physical Data to Forecast Land Use Change and Environmental Consequences for California's Coastal Watersheds.EPA Grant Number: R829803
Title: Integrating Economic and Physical Data to Forecast Land Use Change and Environmental Consequences for California's Coastal Watersheds.
Investigators: Merenlender, Adina , Berck, Peter , Biging, Greg , Landis, John
Current Investigators: Merenlender, Adina , Biging, Greg , Landis, John
Institution: University of California - Berkeley
EPA Project Officer: Michaud, Jayne
Project Period: July 1, 2002 through June 30, 2004
Project Period Covered by this Report: July 1, 2003 through June 30, 2004
Project Amount: $259,454
RFA: Futures: Research in Socio-Economics (2001) RFA Text
Research Category: Nanotechnology , Economics and Decision Sciences , Water and Watersheds
The overall objective of this research project is to examine the environmental consequences of land use change for California’s coastal watersheds, which are experiencing rapid urban and agricultural expansion. These foremost land use stresses can result in cumulative impacts to coastal watersheds that effect anadromous fish. The specific objectives of this research project are to:
- develop a spatially explicit economic land use change model;
- compare the proposed economic modeling approach to a more traditional non-economic (or reduced-form) land use model;
- determine changes in land cover based on the resulting scenarios of land use change;
- and address consequences for coastal Mediterranean watersheds and instream habitats for endangered salmon.
Our approach is to incorporate land valuation into spatially explicit land use change models to predict probabilities of vineyard and residential conversion at the parcel level in Sonoma County, California. Specifically, we propose to develop two economic land use change models, one based on hedonic price techniques and the other on expected net present value (NPV), and then compare them to a more traditional non-economic model based on physical attributes. The third step determines changes in land cover based on the resulting scenarios of land use change. The last step combines these various scenarios of land use change with biophysical models to determine the influence of land use on stream habitat, cumulative impacts, and other potential environmental costs. We draw on existing habitat data at the reach scale from field surveys by the California Department of Fish and Game (CDFG).
Objective 1: Develop a Spatially Explicit Economic Land Use Change Model
Year 1 of the project was spent collecting economic data from the Sonoma County Tax Assessor, compiling appropriate geographic information systems (GIS) layers and developing the spatially explicit economic land use change models. The Sonoma County Tax Assessor’s database provides the necessary information on individual parcels for the land value, current land use, lot size, and many other property characteristics. Once compiled, the Sonoma County Tax Assessor’s complete database was linked to the parcel map via each parcel’s unique identification number. The analysis uses only the parcels located outside incorporated cities, with lot size greater than 0.25 acres, and classified as one of the three main land uses (residential, vineyard, extensive use). The parcel boundaries permitted the overlay and extraction with GIS layers to obtain many site and neighborhood characteristics on land quality, travel times to urban centers, zoning, and other factors. Site-specific characteristics important for either agricultural or residential land values, including land quality (slope, elevation, farmland type, microclimate, 100-year floodplain), accessibility (travel times to urban areas), and zoning (land designations, lot size), were derived from existing spatial data.
David Newburn, a recent graduate supported by this project, helped develop two economic models as part of his thesis. The primary model used hedonic analyses on recent sales of “extensive use” parcels unrestricted from future development to estimate residential and vineyard values. In the first stage, land values for residential transactions were determined as function of the site characteristics and recent residential transactions. The expected residential land value (Vr) could be predicted for each developable parcel, because the site characteristics are known for all parcels. The expected vineyard land value (Vv) could be determined similarly from recent vineyard transactions. In the second stage, we employed a multinomial discrete choice model using recent land use transitions and extensive, residential, and vineyard land use to produce a probability map of conversion. For example, parcels with a high likelihood of residential conversion were expected to have high values for Vr and Vr-Vv.
The second model used a similar model framework, but utilized expected NPV as an alternative economic measure of vineyard land value. This alternative measure was applied because the expected stream of discounted agricultural returns is an underlying determinant of farmland value. Expected NPV could be determined directly from farm level information on grape price, yield, and establishment costs as well as physical attributes. In 2001, we conducted a cross sectional survey, in close collaboration with the Sonoma County Grape Growers Association, to obtain these farm level data. In total, we obtained 207 responses that included spatially explicit data on yield, grape price, percent of parcel that is vineyard, and other site characteristics.
Results to Date . Results from the modeling exercises showed that slope had a significant negative effect on land value, whereas distance to Santa Rosa and growing degree days had a positive effect on land value. Surprisingly, lot size had a significant negative effect on land residential value. For vineyard land value, floodplain, slope, and lot size had a negative effect on land value, whereas grape price and lag in land price positively influenced vineyard land value. Significant lot size and zoning coefficients indicated that current or potential residential use can influence vineyard parcel values. In total, these models could explain 67 percent and 61 percent of the total variation in land price for residential and vineyard land values, respectively.
The expected NPV model focused entirely on the land value from vineyard production. Tobit regression results indicated that the ratio of planted vineyard acres to lot size depended heavily on physical factors such as terrain, microclimate, and soil suitability. Flatter slopes and warmer growing degree day conditions both increased the degree of vineyard planted for a given site. In the second stage of model development, the estimated wine grape revenue for planted acres was determined with respect to physical and management characteristics from the grape growing survey. A negative coefficient for slope indicated that hillside vineyards have a reduction in the expected grape revenues. This turns out to be largely because of decrease in yield with an increase in slope; hillside vineyards continue to receive a grape price premium.
Estimation results from the two structural models demonstrated that higher residential and vineyard land values are positively associated with residential and vineyard conversion. In other words, the land values as explanatory variables help to determine whether a developable parcel was converted, but does not discriminate well between residential and vineyard conversion. The expected NPV model improves the prediction of vineyard conversion by removing the influence of residential potential.
Objective 2: Compare the Proposed Economic Modeling Approach to a More Traditional Non-Economic (or Reduced-Form) Land Use Model
We compared the two economic modeling approaches to a reduced or physically based land use model. For this reduced model, we used recent land use transitions to predict three possible terminal land use states similar to the structural models described above: residential, vineyard, or extensive use. Site characteristics served as the explanatory variables, which determined the underlying suitability under each alternative land use.
For this model comparison, we assessed the accuracy of predicted land use transitions using the same data set. The total population of the 1990 developable parcels was stratified into three groups based on 1990-2000 land use transitions and sampled randomly to provide 500 parcels from each land use state. Predicted conversion probabilities were calculated based on site characteristics for the developable parcels and estimated logistic regression coefficients from each model. The maximum predicted probability among the three states for a given site was taken as the predicted land use transition.
Results to Date . The reduced-form model results in the highest accuracy in predicting land use change, 73.7 percent overall for the three types. The primary structural model has a much lower overall accuracy at 61.1 percent, primarily because of poor prediction of vineyard conversion (42%). The second structural model improves this prediction for vineyard conversion to 70 percent. Although the reduced-form model yields greater accuracy for predicting past land use conversions, it does not explicitly establish a relationship between relative land value and land use conversion. For this reason, the structural models hold more promise in understanding where and when land use conversion will take place.
Objective 3: Determine Changes in Land Cover Based on the Resulting Scenarios of Land Use Change
The land use change scenarios rely on two submodels; regional demand and spatial allocation. Regional demand models are done separately for the housing units and new vineyard acreage estimates. The spatial allocation models are the probabilistic land use change models described in Objectives 1 and 2. For example, regional housing demand is calibrated based on quarterly housing construction data from 1990-2003 (64 observations). Time series analysis is utilized to forecast the expected annual housing units during the time period 2004-2020. These new homes then are allocated stochastically, according to the residential conversion probabilities estimated in spatial land use change models. Policy scenarios are implemented by changing the spatial allocation rules, which guides the distribution of future conversion events.
One of the major advancements in 2004 was the calibration of land use change models to differentiate among residential densities. In Sonoma County, the extent of low-density development (1 unit per 1-5 acres) and very low-density development (1 unit per 5-40 acres) is several times larger than the urban footprint (>1 unit per acre). In the last decade, 85 percent of new homes have been built as dense subdivisions in existing cities or within urban growth boundaries. The rural residential growth, however, has a much larger area, and it often occurs in rugged terrain with unpaved roads. The relative impacts of urban versus exurban residents for sediment production or habitat impacts are valuable for policy makers. Initially, the forecasts from land use change models are done as a “development as usual.” We are able to see how the current regulation regime, such as County General Plans, can actually encourage low-density sprawl. Then, we compare a series of land use change scenarios to determine alternative patterns of land use change and assess the relative environmental impacts from urban, exurban, and agricultural uses.
The following are the different land use change scenarios to be implemented:
- development as usual;
- development as usual plus wine price slump;
- development as usual plus wine price surge;
- maximum environmental protection(policy changes on vineyard growth on hillslopes and streams setbacks);
- and smart city growth (policy changes on rural residential growth).
These land use scenarios represent different states of the regional economy, current, and alternative regulatory regimes. Presently, we are preparing papers for publication based on these forecasts.
Objective 4: Address Consequences for Coastal Mediterranean Watersheds and Instream Habitat for Endangered Salmon
Research projects related to the effects of land use conversion on instream habitat quality are in progress. In particular, we are focusing our efforts on evaluating the impacts of land use changes on sedimentation production to streams and stream hydrology.
In Year 2 of the project, we developed and estimated models to predict the effects of residential and vineyard conversion on instream processes. Jeff Opperman was supported by this grant as a graduate student working on the initial investigations of the effects of land use changes on instream habitat data collected from the CDFG. Instream data were assembled from the CDFG, and most of the data were dynamically segmented and geocoded with the help of Colin Brooks, a GIS analyst partially funded on this project. Kathleen Lohse, a post-doctoral fellow funded through this project, assessed and improved this analysis by incorporating variables that explained additional variation in the stream habitat data.
In this research, we delineated approximately 150 watersheds in the Russian River Basin and extracted watershed and land use change metrics that were potentially important in determining instream and watershed scale responses. In particular, we used land classifications based on LANDSAT Thematic Mapper imagery to determine the proportion of watersheds covered in different land uses. We then focused our analyses on the relationship between land use/land cover and concentrations of fine sediment, termed here embeddedness, found in depositional stream reaches that are potential spawning habitats for salmonids. Furthermore, we examined the scale of influence of land use cover within watersheds as well as across watersheds.
Results from our analyses showed a strong relationship between embeddedness and proportion of watersheds in urban and agricultural land use. The power of the empirical regression model depended on the size of the watershed. In general, the watershed scale was the best predictor of embeddedness compared to other local or drainage network scales of influence. A paper was recently submitted for publication on this research.
To couple this empirical regression stream habitat model to the land use change models, we developed several land use/land cover maps based on parcel level data and tax assessor data rather than LANDSAT imagery data (as done above). These maps resolve some issues of spatial and temporal time scale and boundary conditions. Most of all, these maps solve resolution issues. Typically, geographical scale differs between ecological and economic land use change models. Land parcels modeled in land use change models will not correspond to cell boundaries in an ecological model. Developing parcel-level land use maps also allows us to capture individual decisions and detect the effects of different residential densities (low, medium, and high) on stream conditions. Thus, we are able to identify the relative impacts of different forms of residential housing and land uses (vineyard). Our preliminary results show that low-medium residential housing and agriculture have a greater impact on the concentration of fine sediment in streams than high density housing. Currently, we are in the process of coupling this model to the land use change model scenarios. We are preparing a paper for publication on this research.
We will investigate the effects of vineyard expansion on water demand and supply in tributaries in the Russian River Basin. We hypothesize that salmon are limited by water during the dry season. The expansion of hillside vineyards has placed increased pressure on water resources through diversions and storage offstream (direct removal from a stream during the summer or winter) or groundwater pumping in the uplands. Despite this apparent increasing anthropogenic demand for water, little is known about how these human stream alterations influence high and low seasonal streamflow. Understanding the impacts of land use change on seasonal low flow may be critical for the survival and recovery of federally threatened salmonids in this basin.
For this purpose, we have started to compile data on diversions within Sonoma and Mendocino Counties from the California State Water Resources Control Board (SWRCB). These data include the allocated water in cubic feet per second (cfs), timing of removal, date of approval of water right, a unique identification number, and other relevant data. After a year of the SWRCB failing to deliver a GIS layer with spatial locations, we have now started to digitize these data by hand. This task should be done in the summer of 2004. Once we have digitized these data, we can link these data to our database, determine the number of diversions and storage units required per vineyard acre, and start to predict the potential impacts of this demand on water supply to salmonids. We can also use the land parcel map data to begin to determine the number of wells associated with urban and suburban residential expansion as well as those associated with vineyards. This information on current demand for water associated with land use is extremely valuable.
We also are conducting a more detailed study to investigate the interactions between surface water diversions and stream hydrology in two streams in a subcatchment of the Russian River Basin. These streams are unique in that Maacama Creek in eastern Sonoma County was gauged by the U.S. Geological Survey (USGS) from 1961 to 1982 at a point where it drains a 43-square mile catchment; and Franz Creek, a tributary to Maacama Creek, was gauged by USGS from 1963 to 1968, where it drains a 15.7-square mile catchment. Aerial photograph analysis indicates that streamflow data was collected by the USGS during a period of low development in the watershed; therefore the hydrologic data recorded may be considered undisturbed compared to hydrologic conditions today.
Water rights records may not adequately express the influence of small-scale water management practices on streamflow: stream diversions may operate illegally, resulting in greater flow diverted than records would indicate. Aerial photographs can be used to derive a method to quantify the potential water removed from streams over time. Within these two watersheds, we traced vineyard cover onto a base GIS map using aerial photographs from the 1960s, 1980s, 1990s, and 2000 to illustrate the time-sequence change in land use in the watershed. We can use the extent of vineyards above the Maacama and Franz Creek gauges in 2000 to suggest a maximum amount of water that would be removed from the drainage network. To assess the impacts of vineyards on streamflow, we can use daily per-acre estimates that farmers would put on crops to meet the seasonal needs and then multiply the per-acre water demand by the total acreage of agriculture upstream, creating a maximum estimate of water removed to meet human demand on any day. We can develop a diversion hydrograph as described above and compare diversion values to amounts of water present in a typical wet or typical dry year. Because this process assumes that all agricultural needs are met through direct channel diversion, the calculated daily rates of diversion will overestimate water demand from the stream; some water is acquired through groundwater pumps or from surface water storage during winter.
In collaboration with Drs. Matt Kondolf and Vince Resh, we also installed
instruments to monitor streamflow at eight locations within the Maacama and
Franz Creek watersheds, including both former USGS gauge locations, to quantify
these impacts at shorter time scales. Instruments were installed at locations
where streams are of different size for two purposes:
(1) by varying the size of streams and watersheds they drain, we can study how water moves through the drainage basin between and during storms throughout the rainy season and how long streamflow persists in smaller streams during summer; and (2) by placing stream gauges at locations with varying degrees of upstream vineyard development, we can examine streamflow patterns to identify whether water levels significantly change as a result of water management practices. Much of this field work over the past year has involved installing instruments at appropriate locations (seven of eight being on private property, additionally requiring me to secure access to each location) and beginning to develop rating curves to make the depth measurements recorded by pressure transducers to correspond to streamflow.
Initial analysis of the data indicate two particular findings of interest. First, all streams respond to rainfall events quickly. Mean daily flow (the form of data most commonly available by USGS) does not adequately describe peak flow conditions following rainfall events but in general, is sufficient for describing flow in the days following a rainfall event and the total volume of water to pass the point each day. The preliminary results of depth against time recorded by the stream gauges in winter 2004 were presented in a poster titled “Suitability of Mean Daily Values for Characterizing Stream Flow in Small Streams of Varying Size in Northern Coastal California” at the North American Benthological Society Conference, June 2004. Data collected in the past 3 months also indicate that water levels in streams with vineyards upstream have a tendency to decline suddenly in each day when air temperatures drop to near 32 ° F. If this is an indication of the influence of frost protection on streamflow, then water management practices are capable of causing streamflow to decline by up to 75 percent. The influence of water management practices will become more lucid once rating curves are complete and streamflow values have been assessed for the entire year. We will continue to monitor streamflow through summer to see whether flow declines any sooner at the former Franz Creek and Maacama Creek gauges than historical records indicate.
To evaluate potential impacts of delocalized water management on stream hydrology, we will use the following method. In addition to determining water demand using aerial photos, diversion records maintained by the California SWRCB will be used to determine instream demand. To receive a water appropriation, potential users have been required to file paperwork with the SWRCB stating the year for the diversion to begin, the total amount of water to be removed, the rate of diversion, the time during the year when water may be removed, and the location of the diversion. Since 1970, 10 requests have been made to divert water from Franz Creek, and 35 have been made to divert from Maacama Creek. Each set of records (those for Franz Creek and for Maacama Creek) can be compiled to create a graph of maximum water extraction above the stream gauge permitted for each day throughout the year. Though this graph would not necessarily represent the total amount of water removed through the year (because diversions may not operate every day or throughout the day), it does describe the maximum amount of water permitted for removal on each day throughout the year. The potential effects of these diversions on stream hydrology in Franz and Maacama Creeks are described in a paper titled, “Evaluating Effects of Water Rights Diversions in Coastal California Streams Over Spatial and Temporal Scales,” soon to appear in the Proceedings of the Symposium on Arid Lands (ASCE EWRI 2004). Results from this method show that because of the seasonal hydrology in this region, little streamflow occurs under natural conditions from April through October. If the meteorological conditions that occurred to produce the wet year 1967 were to occur again under the current water management regime, diversions during spring for frost protection would have the potential to reduce discharge by 10 percent in Franz and Maacama Creek. If a dry year such as 1964 were to be repeated today, the current water management regime would result in both Franz and Maacama Creeks going dry.
We also are trying to link changes in land use more directly to fish population dynamics. Kathleen Lohse is currently collaborating with a consulting group from Sonoma County who monitored salmonids for 10 years in seven streams in Sonoma County at multiple spatial and temporal scales, including these historically gauged streams. Preliminary findings from their study show that population dynamics vary with spring streamflow and appear to vary with land use. In particular, water diversions resulting from agriculture appear to be impacting the distribution and access of salmonids to different streams. We currently are linking these data to watershed metrics and land use activities associated with these streams.
Journal Articles on this Report : 3 Displayed | Download in RIS Format
|Other project views:||All 26 publications||3 publications in selected types||All 3 journal articles|
||Newburn D, Reed S, Berck P, Merenlender A. Economics and land-use change in prioritizing private land conservation. Conservation Biology 2005;19(5):1411-1420.||
||Newburn DA, Berck P, Merenlender AM. Habitat and open space at risk of land-use conversion: targeting strategies for land conservation. American Journal of Agricultural Economics 2006;88(1):28-42.||
||Opperman JJ, Lohse KA, Brooks C, Kelly NM, Merenlender AM. Influence of land use on fine sediment in salmonid spawning gravels within the Russian River Basin, California. Canadian Journal of Fisheries and Aquatic Sciences 2005;62(12):2740-2751.||