Project Team Leaders

Dr. Zhihao Wang, Prof. Greg Pasternack, Prof. Yufang Jin

Project Organizations

University of California at Davis, San Jose State University, Contra Costa Resource Conservation District, Napa Resource Conservation District, North Santa Clara Resource Conservation District, SafeR3, University of California Agriculture and Natural Resources

Project Sponsors

This research was supported by funds from the Climate Action 2023 Seed Awards of the University of California, Grant Number R02CP6967. This project was also supported by the USDA National Institute of Food and Agriculture, Hatch project numbers CA-D-LAW-7034-H and CA-DLAW-2620-H.

Introduction

A classification organizes and prioritizes information so that like entities can be matched up and addressed efficiently with a common framework, while unlike entities can be appropriately treated differently. Classifications are often used globally for regional environmental management (Lane et al., 2017). They are also the basis for California’s regional environmental flows framework (Lane et al., 2022).

While stream classification has been pursued by geoscientists and ecologists for more than 70 years and is well justified, there has been a strong bias against considering where artificial streams fit into such classifications. There are a few classifications that consider some types of artificial channels in individual, narrow enigneering contexts (e.g. mining channels, agricultural canals) and several "evolutionary" frameworks that position some types of human-degraded streams into a system. However, there are no geophysical stream classifications that span the diversity of both natural and artificial stream types, let alone one made on an evidence-basis justified by data. 

In this project, we developed a stream classification spanning natural and artificial stream types for the 9-county San Francisco Bay Area. Once the stream types were determined, then we created a set of archetypes (sensu Cullum et al., 2016) to study the details of each stream type. These archetypes can then be used in adaptive management to identify pathways to achieve climate resilience. Through transdisciplinary engagement with archetypes, policy and management recommendations for USC types can be arrived at by consensus, including design guidance for NbS that benefit the most climate-vulnerable urban dwellers. Our activities will help urban & climate planners, social service providers, & homeless-advocacy groups reduce negative impacts of unhoused living on the people experiencing it, while conversely helping reduce negative impacts of unhoused living on USCs & adjacent communities.

Study Area

The study area for this project was the 9-county San Francisco Bay Area.

Experimental Design

Sufficient abundance and variety of field site sampling are crucial for obtaining an accurate reach-scale river classification of a regional stream network in support of scientific research and river management. However, many studies still randomly select field sites or only visit accessible streams. Machine learning has been recognized for discovering and extracting streams' geomorphic patterns efficiently and accurately from data, but its application in field site sampling design is still in its infancy. This study developed a general and practical field site selection framework by incorporating machine learning in a human-in-the-loop manner. This framework includes three steps: (1) initial field site selection via machine learning from prior datasets, (2) selected field site accessibility evaluation and observation, and (3) additional field site decision and selection via an iterative learning process. In an example application to the San Francisco Bay Area (California, USA), our framework extracted representative geomorphic characteristics of both previous stream types from prior labeled and geospatial datasets and previously unrecognized stream types based on uncertainty information obtained by machine learning. Moreover, we proposed methods for replacing inaccessible sites to ensure sufficient information is retained in the selected field sites. The feasibility of the framework was assessed by statistical analyses of results and empirical validation via field surveys.

Details about the protocol are provided in the video at the link below: 

• https://t.ly/DjwjK

A scientific journal manuscript explaining this novel approach to site selection is currently in peer review. A citiation will be provided here when it is published.

Field Data Collection With The PURE Protocol

In this study we developed and implemented a novel fluvial geomorphology, cross-section based topographic surveying methodology that we call the “Protocol for Unbiased River Evaluation” (PURE). Admittedly, nothing is truly free of bias, but the goal of the protocol is to at least eliminate what has turned into a powerful bias in fluvial geomorphic field surveying- the reliance on bankfull determination.

The PURE 1.0 protocol has 5 steps: general site assessment, locating transects, surveying transects, photo documentation, and ancilliary data collection. In our study, the ancilliary data collected was a suite of riverbed sediment grain size distribution metrics.

Details about the protocol are provided in the video at the link below:

• https://t.ly/YLCoE

Site Variables Produced From Field Data

With the data collected using the PURE Protocol, we were able to compute values for up to 93 variables for each site. Variables are organized into five categories: channel morphometry variables, grain-size variables, valley-floor variables, engineering variables documenting human modifications using shape and composition observations, and single- or multi-threaded channel.

Stream Classification R Markdown Program

To develop the stream classification from the variables computed using field data, we philosophically used a "bottom-up" approach. This means that instead of dictating what types would exist a priori, we let the data speak for itself and produce an evidence-based, objectively defensible set of groups. Of course, any typology can be assailed on a variety of rational bases, but our approach conforms to modern geophysical scientific norms. In fact, we used an existing framework and set of tools already peer reviewed and published in the scientific literature in two journal articles. In this framework, we use an R markdown program we made for statewide, natural river typing. We have updated and modified it as needed for this project. This program performs thorough multivariate statistical analyses, including adherence to statistical and physical assumptions and constraints, yielding a regional stream typology, an uncertainty assessment of classification performance, and identification of which variables are driving what aspects of USC typology (Byrne et al., 2020). Essential multivariate methods include non-metric multidimensional scaling, hierarchical clustering using Ward’s algorithm, and classification and regression tree (CART) analysis.

Stream Classification Results Overview

The image below shows the Ward's diagram of the stream classification using a preliminary numbering scheme for each group. The group #s have no meaning per se and are not used as class names- they are just part of the R Markdown program approach. because there are 164 sites used to make the classification, it is difficult to see the details of this diagram, but the main point is that we used a rational, objective approach and obtained defensible results in support of distinguishing 12 stream types, of which 6 are artificial and 6 are natural.

The CART analysis was able to correctly classify 95% of sites. With robust cross-validation testing, classificaiton accuracy was 83 %, which is high for this type of analysis and data.

A video presentation explaining our classification methods and results is pending recording.

A scientific journal manuscript has been drafted and is currently going through internal review and editing by our co-author team in advance of submission to scientific peer review.

Key Concepts To Understand The Classification

What is an "artificial" or a "natural" stream? We have produced a draft report explaining these concepts and illustrating them with several examples.

• https://drive.google.com/file/d/1ybqtjp_jInTb039ZAqfRTNUdvtud_xlO/view?usp=sharing

What do you need to know to successfully employ this classification system to differentiate SFD Bay Area Stream types? A video presentation is in preparation (will be linked below when ready). Here is the link to a PDF of the slide deck.

• https://drive.google.com/file/d/1vvHF2PE8g3WhmZYeYpWr5ciaA24zEqau/view?usp=sharing

Introducing the SF Bay Area Stream Types

The following columns show the 12 stream types. you can click on them to see detailed information about each type. if nothing happens when you click an image or hyperlink, it is because we have not yet populated that part of the website yet.

Dichotomous Key Stream Typing App

Fortuitously, the objective CART analysis yield in an approach to determinig which stream type is present at any site that can be achieved entirely by visual inspection. No quantitative measurements are needed.

That said, one does need some basic training in the concepts that underpin the classification to be able to correctly identify the different situations present.

Also, the classification applies to 200-m stream intervals, but typically people cannot observe a whole 200-m site safely on a reocnnaissance basis. Safety and respecting private property are important. Therefore, care is needed when applying the classification to a site on the basis of a limited view of a site.

With practice, both professional and citizen scientists can accurately use the classification.

Help Us Train Our ML Model

We have a machine learning model that predicts the stream type for any of the 42,615 individual 200-m stream intervals in the study area. Currently, it is trained with several hundred sites. The more stream-type labels we can get, the more accurately we can predict stream types for the remaining stream intervals and accurately characterize the uncertainty in those predictions.

Resources to aid stream typing

First, refer to the files and videos linked above on the page.

Second, the dichotomous key is provided in a slidedeck and video presentation at the links below. because the phone app does not allow images, it i svery helpful to have this slide deck handy on your device or printed out to refer to.

• Slide deck: https://drive.google.com/file/d/1Ovqde7_Q9JXOPuf9iOJGFn-3PS5vlJZZ/view?usp=sharing
• Video: (pending)

ArcGIS Survey123 App

Anyone may use the S.F. Bay Area Stream Typing survey via the free app, Survey123.

While it is not required to create an account to use the app, it helps if you do, so that we can know something about who is using the app to help evaluate data accuracy. We do accept anonymous submissions.

• Link to Survey123 App survey: (pending)