LANL COVID-19 Cases and Deaths Forecasts

Authors

• Lauren A. Castro
• Geoffrey Fairchild
• Isaac Michaud
• Dave Osthus

Description

This repository contains an archive of the forecasts published on https://covid-19.bsvgateway.org/ from 2020-04-05 until 2021-09-27. Forecasts were published every Monday and Thursday unless otherwise noted in Key Dates. Forecasts published on Mondays used observed data through Sunday; forecasts published on Thursdays used observed data through Wednesday.

All forecasted quantities were saved as csv files and conform loosely to the standards set by https://covid19forecasthub.org/. Each file contains historical observations starting with 2020-01-22 and ending on the last observed date and the forecasts for the next 48 days or 6 weeks. All forecast files names are prefixed with the last observed date used in the forecast. For example, a forecast published on the website on 2021-09-20 (a Monday) would have 2021-09-19 in its file name.

A metadata file (forecast_metadata.json) is also provided for ease of searching.

Over the course of 18 months, we implemented two different forecasting models. We switched to Version 2 (V2) of our forecasting model on 2020-10-28.

Version 1: April 5, 2020 - October 25, 2020

Methodology

The V1 model was a combination of a compartmental susceptible-infectious model and a hierarchical, Bayesian model. The model captured how people moved from susceptible to confirmed cases to deaths. The geography-specific growth rate parameters (i.e., how quickly the confirmed cases grew/declined) and the geography-specific case fatality fractions (i.e., the percentage of confirmed cases that resulted in death) were allowed to vary with time and were modeled hierarchically across neighboring geographies; they were learned from data rather than hard-coded. The main assumption of the V1 model was that the growth rates would tend towards zero over time as a result of sustained mitigation efforts.

Forecast Files

Five forecast types were produced using V1 of the forecasting model at the US and global levels. See Key Dates for exact start and end dates for the confirmed_cases_peak_timing and *_incidence_* forecasts.

Name	Time Frame	Forecast Horizon	Description
confirmed_incidence_quantiles	daily	48 days	forecasted daily confirmed cases
confirmed_quantiles	daily	48 days	forecasted cumulative confirmed cases
deaths_incidence_quantiles	daily	48 days	forecasted daily deaths
deaths_quantiles	daily	48 days	forecasted cumulative deaths
confirmed_cases_peak_timing	weekly	6 weeks	cumulative probability of peak occuring in week of weekdate

The full names of the output CSVs from V1 are in the following format: -- fcstdate is the forecast date (e.g., 2020-10-27) -- target can be cases or deaths -- type can be incidence or blank for cumulative -- quantiles indicates the forecast is summarized into quantiles -- geography can be us or weekly -- website indicates the forecast was published. Note, early forecast CSVs are missing this tag.

Output Column Descriptions

There are 23 quantiles forecasted for each quantity. The quantile columns are names "q.*" where * is the quantile. All of the names are:

 [1] "q.01"  "q.025" "q.05"  "q.10"  "q.15"  "q.20"  "q.25"  "q.30"  "q.35"  "q.40"  "q.45"  "q.50"  "q.55"  "q.60"  "q.65"
[16] "q.70"  "q.75"  "q.80"  "q.85"  "q.90"  "q.95"  "q.975" "q.99"

V1 files have additional columns which identify the locale and date of either the forecast or historical observation.

Column name	Dtype	Description	Range or Example	Notes
dates	string	last observed date	"2020-04-01" through current	All observations on and before fcst_date are used for make forecasts
simple_state	string	unique id for each locale	e.g. "newmexico"	Lower case state or country name with spaces and punctuation removed
state	string	Name of locale as reported by JHU CSSE	e.g. "New Mexico"	Includes capitalization, spaces, and punctuation
obs	bool	indicator of whether the entry was observed	0,1	All entries dated on or before fcst_date are observed
truth_confirmed	integer	reported number of confirmed cases	[0,∞)	Applies to cumulative and incident records. Forecasted entries have NA.
truth_deaths	integer	reported number of deaths	[0,∞)	Applies to cumulative and incident records. Forecasted entries have NA.
fcst_date	string	last observed date	"2020-04-01" through current	All observations on and before fcst_date are used for make forecasts
big_group	string	WHO global regions	e.g. "NORTHERN AFRICA AND WESTERN ASIA"	Only occures in global forecasts. Not in US state forecast output files.

Version 2: October 28, 2020 - September 27, 2021

Methodology

A description of V2 of the forecasting model, COVID-19 Forecasts using Fast Evaluations and Estimation (COFFEE), can be accessed here.

Forecast Files

Eight types of forecasts were produced using V2 of the forecasting model. Daily forecasts were made for the 48 days following the last observed day. Weekly forecasts were made for the 6 following epiweeks after the current observed epiweek. Because the COVID-19 Forecast Hub requests forecasts for the first full epiweek (except for Mondays which include data observed the previous Sunday) any forecast made with data ending on any day other than Saturday or Sunday will skip the current epiweek and aggregate daily forecasts beginning the following unobserved Sunday. This skip results in a missing epiweek entry in the weekly aggregation output files for the current epiweek. Also note that because of this definition of epiweek, the COVID-19 Forecast Hub are one week ahead of the CDC epiweek defined in the lubridate package.

Name	Time Frame	Forecast Horizon	Description
incidence_daily_cases	daily	48 days	forecasted daily confirmed cases
cumulative_daily_cases	daily	48 days	forecasted cumulative confirmed cases
incidence_daily_deaths	daily	48 days	forecasted daily deaths
cumulative_daily_deaths	daily	48 days	forecasted cumulative deaths
incidence_weekly_cases	weekly	6 weeks	forecasted weekly confirmed cases
cumulative_weekly_cases	weekly	6 weeks	forecasted weekly cumulative confirmed cases
incidence_weekly_deaths	weekly	6 weeks	forecasted weekly deaths
cumulative_weekly_deaths	weekly	6 weeks	forecasted weekly cumulative deaths

The full names of the output CSVs from V2 are in the following format:

• fcstdate is the forecast date (e.g., 2020-10-27)
• geography can be us or global
• type can be incidence or cumulative
• resolution can be daily or weekly
• target can be cases or deaths
• website indicates the forecast was published

Output Columns

As with V1, there are 23 quantiles forecasted for each quantiy. The quantile columns are names "q.*" where * is the quantile. All of the names are:

 [1] "q.01"  "q.025" "q.05"  "q.10"  "q.15"  "q.20"  "q.25"  "q.30"  "q.35"  "q.40"  "q.45"  "q.50"  "q.55"  "q.60"  "q.65"
[16] "q.70"  "q.75"  "q.80"  "q.85"  "q.90"  "q.95"  "q.975" "q.99"

The remaining columns identify the locale and date of either the forecast or historical observation. Not all columns appear in all output files. For example, instead of a date column weekly_confirmed_incidence_quantiles and weekly_deaths_incidence_quantiles have a start_date and end_date fields that correspond to the beginning and end dates of the observed data used to compute the observed incidence for historical data and to the beginning and end date of the forecasted days used to create the incident forecast. The CDC epiweek of these date ranges is different than the COVID-19 Forecast Hub defined epiweek. For weekly_confirmed_quantiles and weekly_deaths_quantiles, instead of a date column they have an end_date field which describes the last date of the COVID-19 Forecast Hub epiweek (which is always a Saturday) or the forecasted cumulative quantity according to the week-ahead rule used by the COVID-19 Forecast Hub.

Column name	Dtype	Description	Range or Example	Notes
fcst_date	string	last observed date	"2020-04-01" through current	All observations on and before fcst_date are used for make forecasts
date	string	date of the observation or forecast	"2020-01-22" through fcst_date + 48 days
key	string	unique id for each locale	e.g. "newmexico"	Lower case state or country name with spaces and punctuation removed
epiyear	integer	CDC defined epiyear	{2020,2021,...}	The first couple of days of the new year are in the previous epiyear
fcst_hub_epiweek	integer	COVID-19 Forecast Hub defined epiweek	[5,52]	Epiweeks reset on the first Sunday of the new year
start_date	string	first day of epiweek observation or forecast	"2020-04-01" through fcst_date + 48 days	Always a Sunday
end_date	string	last day of epiweek observation or forecast	"2020-04-01" through fcst_date + 48 days	Always a Saturday
week_ahead	integer	number of epiweeks ahead of fcst_date of forecast	{1,2,3,4,5,6}
name	string	Name of locale as reported by JHU CSSE	e.g. "New Mexico"	Includes capitalization, spaces, and punctuation
obs	bool	indicator of whether the entry was observed	0,1	All entries dated on or before fcst_date are observed
truth_confirmed	integer	reported number of confirmed cases	[0,∞)	Applies to cumulative and incident records. Forecasted entries have NA.
truth_deaths	integer	reported number of deaths	[0,∞)	Applies to cumulative and incident records. Forecasted entries have NA.
big_group	string	WHO global regions	e.g. "NORTHERN AFRICA AND WESTERN ASIA"	Only occures in global forecasts. Not in US state forecast output files.

Key Dates

Below we note significant changes to the format or type of output ("Output Change"), or dates when we did not adhere to the normal Monday/Thursday schedule of publishing forecasts due to model or data issues.

Forecast Date	Type	Description	Version
2020-04-05	Output Change	First US daily cumulative forecasts published	V1
2020-04-15	Output Change	First US daily incidence foreasts published	V1
2020-04-26	Output Change	First global daily cumulative and incidence forecasts published	V1
2020-06-14	Model issue	Published US/global output from 2020-06-13	V1
2020-06-24	Data issue	Published US/global output from 2020-06-23	V1
2020-06-24	Output Change	Peak forecast terminated	V1
2020-10-28	Output Change	V2 COFFEE implemented for US and global forecasts	V2
2020-12-23 -- 2021-01-02	Output Change	Pause for winter closure - no forecasts published	V2
2021-01-06	Model issue	Published output from 2021-01-05	V2
2021-03-17	Model issue	No forecasts published	V2
2021-05-30	Model issue	No forecasts published	V2
2021-06-10	Output Change	Thursday forecasts terminated	V2
2021-09-27	Output Change	Last US/global forecasts	V2

Notice and Disclaimer

For Scientific and Technical Information Only

© Copyright Triad National Security, LLC. All Rights Reserved.

For All Information

Unless otherwise indicated, this information has been authored by an employee or employees of the Triad National Security, LLC., operator of the Los Alamos National Laboratory with the U.S. Department of Energy. The U.S. Government has rights to use, reproduce, and distribute this information. The public may copy and use this information without charge, provided that this Notice and any statement of authorship are reproduced on all copies. While every effort has been made to produce valid data, by using this data, User acknowledges that neither the Government nor Triad makes any warranty, express or implied, of either the accuracy or completeness of this information or assumes any liability or responsibility for the use of this information. Additionally, this information is provided solely for research purposes and is not provided for purposes of offering medical advice. Accordingly, the U.S. Government and Triad are not to be liable to any user for any loss or damage, whether in contract, tort (including negligence), breach of statutory duty, or otherwise, even if foreseeable, arising under or in connection with use of or reliance on the content displayed on this site.