Study Area
The study area for the project is the 2,303 coastal communities within Long Island Sound (LIS) watershed - located in four Connecticut (CT) counties (Fairfield, New Haven, Middlesex, and New London), and three New York (NY) counties (Westchester, Nassau, and Suffolk). See the Outreach Targeting Tool for a synoptic view.
Methods Overview
The methodology is outlined in the following steps.
- Conduct Fertilizer Use Survey
- Develop Fertilizer Use Prediction Model
- Calculate nitrogen (N) load from residential lawn fertilizer
- Predict Fertilizer Reduction
- Estimate N delivery to LIS
- Calculate Reduction in N delivery
- Summarize and rank results by neighborhood
Methods
1. Fertilizer Use Survey
Invitations to complete the online survey were sent via postal mail to a random sample of 30,000 single-family residential homes with lot sizes between 0.1 and 5.0 acres in the study area. Each invitation included a unique login ID and password that allowed us to geolocate a survey response to its corresponding parcel. We received 2,344 completed survey responses in CT and NY (7.8% response rate). Survey questions elicited information on lawn and landscape behaviors under current and potential future conditions, including current lawn fertilizer use and willingness to reduce fertilizer use under alternative future scenarios. The survey also elicited information on household demographic information and property characteristics. Data from survey responses was supplemented with information on property characteristics, including house age, house size, lot size, from spatially-explicit parcel datasets with associated tax-assessor data.
Key Findings
The amount of lawn fertilizer applied is correlated with variables, including:
- house size
- house age
- lot size
- household income.
In summary, homeowners residing in newer, larger homes on larger area properties with higher household income often tend to have a higher amount of fertilizer applied.
2. Fertilizer Use Prediction Model
The combined survey and parcel datasets was used to develop statistical models that determine if/how property characteristics and household demographics are related to households’ fertilizer use. The approach is grounded in a two-step model of lawn fertilization decisions:
- Each household makes a binary decision on whether to fertilize (yes or no).
- The household chooses the number of applications conditional on the decision to fertilize.
Both stages were modeled as a function of variables on household demographics and property characteristics. Property characteristics (e.g., house size, house age, lot size) were obtained for each household from the parcel datasets. Household demographics were derived from the U.S. Census American Community Survey (2016–2020) using a simulation approach. See Newburn et al. (2025) for technical details.
The statistical model was then used to predict the number of lawn fertilizer applications for the single-family households with lot sizes from 0.1 to 5 acres in the study area. The single-family households were identified according to residential land-use codes in the parcel datasets.
Model Application
The prediction model was NOT applied to multi-family households because lawn care in multi-family housing is typically managed by property managers, homeowners’ associations, or commercial landscaping firms rather than individual residents. Since our survey targeted single-family households, the resulting fertilizer-use behaviors are not intended to represent multi-family households. Fertilization behavior for multi-family parcels is thus assumed to remain constant within the illustrated scenarios.
Parcel-based Fertilizer N Load
The Nitrogen (N) load from residential lawns fertilizer represents the amount of N applied to the landscape. The N load of each residential parcel was calculated as:
N Load (lbs/year) = Estimated Number of Fertilizer Application (application/year) * N Load Rate (lbs/acre/application) * Lawn Area (acre)
The estimated number of fertilizer applications for each single-family parcel was derived from the fertilizer use prediction model. We assumed that all multi-family households (excluding condos and apartment complexes) fertilizer once a year.
The N load rate for residential lawn fertilizer (43.7 lbs/acre/application) was adopted from the Long Island Sound Nitrogen Loading Model.
We mapped the lawn area per parcel across CT and NY using NOAA’s high-resolution (1-meter) land cover dataset for each state. Because data availability varies by state, we applied two distinct state-specific approaches:
- CT: extracted lawn area directly from 2016 CT 1-meter land cover dataset. The "developed open space" category serves as the proxy for the lawn footprint.
- NY: estimated lawns within residential parcels by subtracting 2021 impervious, water, and forest layers from the total parcel area.
The figures below illustrate the mapped lawn footprints from CT and NY compared to high-resolution aerial imagery.
Calculation Example:
A single-family parcel with 0.25 acres of lawn area was predicted to apply fertilizer twice a year. The fertilizer N load from this parcel was calculated as
N load (lbs/year) = 2 applications/year * 43.7 lbs/acre/application * 0.25 acres = 21.9 lbs/year
Pixel-based Fertilizer N Load
To ensure high spatial accuracy, we mapped and aggregated the parcel-based fertilizer N loads into a N load map at 30-meter pixels using a multi-step spatial allocation process:
- parcel allocation: the calculated N load for each parcel was divided by its total number of 1-meter "lawn" pixels. This average load was then assigned equally to every individual "lawn" pixel within that parcel.
- grid alignment: We generated a standard 30-meter grid to directly align to the N-Sink Transport Index map.
- load aggregation: The final value for each 30-meter grid cell was calculated by summing the N loads of all 1-meter "lawn" pixels nested within that cell.
Fertilizer Reduction Prediction Model
The household survey also collected information about homeowners’ willingness to reduce current fertilizer use in response to alternative types of statewide policies and programs designed to influence household fertilizer use. The resulting "willingness to reduce" data was used to develop statistical models that predict the percent reduction in fertilizer application per single-family parcel due to hypothetical behavior-change campaigns.
The data from behavior-change questions was analyzed using two-step hurdle models, estimated for the subsample of respondents who currently apply fertilizer:
- The first-step logit regression predicts the probability of agreeing to a fertilizer reduction in response to a specific type of behavior-change program (yes or no).
- The second-step fractional logit regression predicts the percentage of reduction, conditional on agreeing to reduce fertilizer use in the first step. The model also included variables on household and parcel characteristics to enable these predictions to vary across household types.
NOTE: For illustration purpose, the hypothetical behavior-change campaign used in the Outreach Targeting Tool assumed no restriction on fertilizer applications, no free lawn assessment, no fertilizer surcharge, no improvement to LIS water quality, no increase in fees as the cost to households, and a 30% reduction in lawn chemical exposure for children and pets.
Parcel-based Fertilizer N Load after reduction
The fertilizer N Load after reduction represents the predicted amount of N after potential reduction in response to a certain scenario. The N load after reduction for each single-family parcel was calculated as:
N Load after reduction (lbs/year) = N Load (lbs/year) * (1- Predicted Percent Reduction (%))
NOTE: Since our survey targeted single-family households and the resulting fertilizer-use behaviors may not represent multi-family households, no potential reductions were assumed for multi-family parcels.
Calculation Example:
A single-family parcel was predicted to have 20 lbs of fertilizer N load per year and be willing to reduce fertilizer use by 25%. The predicted N load after reduction from this parcel was calculated as
N load after reduction = 20 lbs /year * (1- 25%) = 15 lbs/year
Pixel-based Fertilizer N Load after reduction
Using a similar approach to the N load mapping before reduction, the parcel-based N load after reduction was also aggregated into a 30-meter map.
5. N Delivery using N-Sink
The N delivery represents the amount of N reaching LIS through its flow path across the landscape. The N delivery at a 30-meter pixel was calculated as
N Delivery (lbs/year) = N Load (lbs/year) * N-Sink N Transport Index (%)
N-Sink is a tool that uses particle tracking analysis to estimate the N pathway from source to receiving water and calculates cumulative N removal by the landscape sinks (e.g., lakes, ponds, wetlands, forested stream corridors, etc.) along that pathway. N-Sink N Transport Index estimates the percentage of N applied at a 30-meter pixel that is expected to reach the downstream receiving water.
6. Reduction in N Delivery
To highlight the areas that can have the greatest potential impact on reducing N delivery to LIS, the reduction in N delivery was calculated as the difference in N delivery before and after reduction for each 30-meter pixel.
7. Summarize and Rank by Neighborhood
To focus outreach effort on where it matters most, we aggregated the 30-meter maps up to the neighborhood level as represented by census block groups. We then ranked these neighborhoods into five priority tiers (from high to low priority) using the “natural breaks" method. Finally, we analyzed these rankings separately for each state to ensure local teams can easily identify and target their highest-priority communities.
NOTE: Rankings of the neighborhoods that are dominated by non-single-family households (e.g., multi-family households, commercial, industrial) were identified as “Not Predicted” as our survey targeted single-family households and the resulting fertilizer-use behaviors may not represent other property owners.
Calculation Example:
A neighborhood was predicted to apply 800 lbs and 200 lbs of fertilizer N load per year from single-family and multi-family households, respectively. The total residential fertilizer N load for this neighborhood was calculated as
Fertilizer N Load (lbs/year) = 800 + 200 = 1,000 lbs/year
The average N transport Index for this neighborhood was 50%. The residential fertilizer N delivery was calculated as
Fertilizer N Delivery (lbs/year) = 1,000 lbs/year * 50% = 500 lbs/year
On average, single-family households in the neighborhood were predicted to reduce their fertilizer use by 20%. The fertilizer N load after reduction was calculated as
Fertilizer N Load after reduction (lbs/year) = 800 * (1-20%) + 200 = 840 lbs/year
The fertilizer N delivery after reduction was calculated as
Fertilizer N Delivery after reduction (lbs/year) = 840 * 50% = 420 lbs/year
The reduction in N delivery was calculated by
Reduction in N Delivery (lbs/year) = 500 – 420 = 80 lbs/year
Support for this project was provided by the Long Island Sound Partnership, in collaboration with Clark University, University of Maryland, and University of Miami.
