Identifying causal changes in landscape greenness with very high-resolution airborne multispectral imagery and a panel data model

《Identifying causal changes in landscape greenness with very high-resolution airborne multispectral imagery and a panel data model》

打印

作者: Allison Lassiter

来源: URBAN FORESTRY & URBAN GREENING,Vol.67,Issue1,Article 127380

语言: 英文

关键字: Change detection;NDVI;Time-series;Turfgrass;Urban landcover

作者单位: Department of City and Regional Planning, University of Pennsylvania, 210 South 34th Street, Philadelphia, PA, 19104, USA;Department of City and Regional Planning, University of Pennsylvania, 210 South 34th Street, Philadelphia, PA, 19104, USA

摘要: To effectively manage urban environmental systems, interventions on private lands may sometimes be necessary. This is illustrated in many water districts where water agencies run programs that provide incentives to landholders to change their landscaping to meet the water agency’s management goals. It is difficult to monitor, track, and evaluate landscape change programs on private land, however. To most efficiently do so, very high-resolution, time-series measurements of urban landcover would be useful, but such data is often unavailable. This study evaluates how private urban landscape change is represented in the highest resolution, publicly available, multispectral data product distributed in the United States. The study focuses on residential parcels that participated in a lawn replacement rebate program run by the East Bay Municipal Utility District, a water agency in California. Each parcel has a known quantity of turfgrass converted to drought-tolerant landscaping. The study asks if Normalized Differential Vegetation Index (NDVI) values derived from National Agriculture Imagery Program (NAIP) data can be used to detect participation in a lawn replacement program at the parcel-level. While the NAIP data product has substantial error in a time-series and urban context, this study finds that approaching change detection using a fixed effect panel data model accommodates error, making it possible to identify NDVI change. With control groups defined by matching and weighing with Covariate Balancing Propensity Scores, results indicate that observed browning at the parcel-level can be causally attributed to participating in the lawn replacement program. This ability to detect small, causal changes in a heterogenous environment opens possibilities for monitoring and evaluating programs acting on private urban landscapes, contributing to more effective and efficient design, pricing and implementation.