<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Mike Dvorak, PhD on Medium]]></title>
        <description><![CDATA[Stories by Mike Dvorak, PhD on Medium]]></description>
        <link>https://medium.com/@mike-dvorak?source=rss-9713d897e298------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*IXxVLMuGr-TQp0zoQDC1MA.jpeg</url>
            <title>Stories by Mike Dvorak, PhD on Medium</title>
            <link>https://medium.com/@mike-dvorak?source=rss-9713d897e298------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 05:53:01 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mike-dvorak/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Your Land’s Water Story]]></title>
            <link>https://mike-dvorak.medium.com/your-lands-water-story-84f88bac3649?source=rss-9713d897e298------2</link>
            <guid isPermaLink="false">https://medium.com/p/84f88bac3649</guid>
            <category><![CDATA[land-ownership]]></category>
            <category><![CDATA[gardening]]></category>
            <category><![CDATA[weather]]></category>
            <category><![CDATA[ranching]]></category>
            <category><![CDATA[farming]]></category>
            <dc:creator><![CDATA[Mike Dvorak, PhD]]></dc:creator>
            <pubDate>Sun, 28 Jan 2024 01:07:40 GMT</pubDate>
            <atom:updated>2024-01-30T19:01:59.887Z</atom:updated>
            <content:encoded><![CDATA[<h4>Creating a robust annual rainfall measurement for your land is easy and cheap using this citizen-science project</h4><h4>Unravel your land’s water mystery for $48</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UOYsf0dMb7ldjLp2ZgrjsA.jpeg" /><figcaption>A simple and dead-accurate rain gauge to record <strong>exactly</strong> how much precipitation your farm, ranch, or homestead gets on an annual basis. Photo from back deck.</figcaption></figure><p>Do you know <em>exactly </em>how much water your land received last year? Or possibly even more importantly, do you know the average amount of rain your land is <em>likely</em> to receive? Rain averaged over weekly, monthly, and annual time periods plays an oversized role in deciding how we should manage our land.</p><p>No single weather variable is spoke about more in diners, feed stores, and on local news than precipitation. While precipitation can come in the liquid (rain) or solid (snow) variety, annual average rainfall can be normalized by using the liquid equivalent, i.e. inches or centimeters of rain per year.</p><p>Why does knowing exactly how much rain and snow your land received matter so much? Because without knowing your exact rain amounts, you are unable to connect patterns from year-to-year related to vegetation growth which relates to sustenance for your animals, your wildfire risk, your waterway health and productivity, and the ideal timing for prescribed fire, should you choose to put fire on your own land. Further, if you’re like me and harvest your own rainwater, knowing the timing and amount of rainfall helps you utilize as much of that rainwater as possible during the dry periods.</p><h3>A $48 solution to measure your annual rainfall</h3><p>What if I told you that for $48, you could own all the equipment necessary to create your own rain climatology for your land? And you would be helping government agencies and researchers alike study weather and climate in your area? Behold the <a href="https://cocorahs.org/">Community Collaborative Rain, Hail &amp; Snow Network</a> (CoCoRaHS) rain gauge; a simple yet dead-accurate 4-inch diameter rain gauge made of plastic, as shown in the top image of this article. The three-piece gauge contains a graduated cylinder that holds one inch of rain. The entire gauge is capable of holding about 11 inches of rain in total and can even be used to measure snowfall.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UvpFddE0nCRu2PfJmC0JIw.png" /><figcaption>Each day, thousands of CoCoRaHS citizen scientists submit their precipitation reports across the US. Image from <a href="https://maps.cocorahs.org/">https://maps.cocorahs.org/</a></figcaption></figure><p>I think an obvious question that a lot of readers of this article will ask, “Why is a manual rain gauge a more ideal solution with all of the automatic weather stations available?” A CoCoRaHS rain gauge has no moving parts and is exceptionally accurate. Automatic rain gauges, the most common of which is a <a href="https://en.wikipedia.org/wiki/Rain_gauge#Tipping_bucket_rain_gauge">tipping bucket</a>, are prone to underestimating intense rainfalls due to spillage over the bucket side. <a href="https://crimsonpublishers.com/acet/fulltext/ACET.000564.php">One study</a> found a 40% error when the rainfall rate reach 2.6 inches/hour (600 mm/hour). Although this rate of rainfall is rare, it is possible you could be missing several percent of your annual rainfall due to this type of error.</p><p>Another advantage of the simple CoCoRaHS rain gauge is the ability to collect snow using the same instrument. All you need to do to collect snow in your CoCoRaHS gauge is to remove the funnel and graduated cylinder from the main rain gauge body which leaves a 4-inch diameter cylinder. Then you bring the gauge into your house to warm and melt the snow. Most consumer-grade weather stations do not have the ability to measure snow.</p><p>Lastly, checking your CoCoRaHS rain gauge makes you inherently connected to your land. There’s nothing like waking up in the morning and heading straight outside to see how much precipitation the planet has gifted you with. But if you’re gone for a vacation, are an absentee landowner, or simply too busy to check your rain gauge daily, you can simply submit a multi-day report that spans up to 60 days for your station. In the Mediterranean climate of California, you typically do not see rain for months at a time; it is not necessary to submit daily “zero” reports. You simply submit one multi-day report covering up to 60 days with the sum total of precipitation in your gauge.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*_YrjatrZ4sWX3lL21_HQzQ.jpeg" /><figcaption><a href="https://apps.apple.com/us/app/cocorahs-observer/id827714558">iOS</a> and <a href="https://play.google.com/store/apps/details?id=com.appcay.cocorahs&amp;pli=1">Android</a> CoCoRaHS apps makes daily reporting pretty simple and convenient. You can also submit reports via the <a href="https://www.cocorahs.org/">CoCoRaHS website</a> and view your station analytics.</figcaption></figure><h3>Your land climatology from CoCoRaHS for free</h3><p>By maintaining a CoCoRaHS station, you automatically can latch into a beautiful dashboard of your precipitation climatology for free. I have provided a few example years of my annual measurements below, along with commentary in the image captions, to show how the climatology helps me relate to my land.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9mkM5mePSgZ0355rllHn7A.png" /><figcaption>This was the first year I owned my farm and it was a pretty slim water year, with only about 60% of normal rainfall.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*76LSi1rq2i2YDXEdLRAf3w.png" /><figcaption>The 2021–2022 was a pretty normal water year for my property. This was a good benchmark year to see how the creek that runs through my property and fills up the adjacent pond behaves during a “normal” year.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N8NL1qVjXqJcOcObk5Tq4A.png" /><figcaption>The 2022–2023 was a crazy water year for much of California. We got two seasons’ worth of rain in a single season. Truly remarkable.</figcaption></figure><h3>What can you do with your rain climatology?</h3><p>To give you an idea of the power of maintaining your own precipitation climatology, I have included a rainwater catchment system that I recently designed and built on my farm. Using my CoCoRaHS station history and some simple Python code in a Jutpyer Notebook, I used my daily precipitation reports for my farm to simulate how much water I could obtain from a rainwater tank connected to my home-roof gutters.</p><p>The daily observations of rain from my CoCoRaHS gauge helped me pick the right size rainwater catchment storage tank and completely eliminated any guessing about how much rain my land <em>actually</em> received. Because I knew exactly on what days the rain came, I did not have to make any simplifying assumptions related to splitting daily rainfall across monthly averages. As I expand my rainwater catchment storage capacity on my land, I will continue to utilize my CoCoRaHS dataset to operate my rainwater catchment system more like a hydraulic-dam operator; the CoCoRaHS dataset makes this possible.</p><p>In my rainwater catchment simulation, I calculated or assumed for my farm:</p><ul><li>One inch of rain on the roof of my house equates to <em>2,207 gallons </em>(!) of water collected</li><li>Daily use of the collected rainwater on a 25&#39; x 50&#39; gardening plot at one inch of water per week or 110 gallons/day</li><li>Garden watering only occurs between May and October</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/891/1*Svr-U0uFzOuNx73D95x0MA.png" /><figcaption>Rainwater catchment simulation for my farm using the daily precipitation totals from my CoCoRaHS rain gauge data. This simulation assumes a 5,000 gallon tank, my roof which is 3540 ft², and about 110 gallons/day use during the gardening season to water a 25&#39;x50&#39; gardening plot with one inch of water per week.</figcaption></figure><p>As you can see in the plot above, in the Mediterranean climate I live in, water is either a feast or famine throughout the year. Basically one winter storm completely fills the rainwater catchment tank with only 2–3 inches of rain. And since I am not gardening in the winter, there is little demand for the collected rainwater. To increase the utilization of my collected rainwater, I have started water wintertime cover crops that still need water due to poor soil conditions. The CoCoRaHS dataset for my farm made all of this analysis and insight possible.</p><h3>Getting started with CoCoRaHS in 3 easy steps</h3><p>You can start building your own rain dataset for your own farm, ranch, or homestead for $48. It is as easy as:</p><ol><li>Buy and install an approved CoCoRaHS rain gauge (e.g., <a href="https://weatheryourway.com/products/official-cocorahs-gauge">this gauge</a>)</li><li>Apply to be a<a href="https://www.cocorahs.org/Application.aspx"> CoCoRaHS observer</a></li><li>Start making observations</li></ol><p>There is no other cost associated with creating a reference-grade rain climatology for your land. And you’re contributing to a citizen science project that helps everyone in your community have more accurate information about their weather and climate. Here are some outstanding <a href="https://www.youtube.com/playlist?list=PL86DC4C330F518387">CoCoRaHS YouTube training videos</a> to help you get you started.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=84f88bac3649" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Enphase Encharge home batteries left us without water and power]]></title>
            <link>https://mike-dvorak.medium.com/enphase-encharge-home-batteries-left-us-without-water-and-power-c71d04753439?source=rss-9713d897e298------2</link>
            <guid isPermaLink="false">https://medium.com/p/c71d04753439</guid>
            <category><![CDATA[batteries-and-chargers]]></category>
            <category><![CDATA[well-water-pump]]></category>
            <category><![CDATA[preparedness]]></category>
            <category><![CDATA[solar-energy]]></category>
            <category><![CDATA[renewable-energy]]></category>
            <dc:creator><![CDATA[Mike Dvorak, PhD]]></dc:creator>
            <pubDate>Thu, 25 May 2023 17:57:40 GMT</pubDate>
            <atom:updated>2023-10-20T17:22:37.831Z</atom:updated>
            <content:encoded><![CDATA[<h4>A $20,000 set of Enphase Encharge home batteries that are supposed to charge from our solar panels died so hard when our grid power went out, we were left without water and power. And we needed <strong>*grid power</strong>* to restart the stone-cold-dead Enphase batteries.</h4><h4>Enphase batteries should not be eligible for reliability subsidies until a fatal flaw is fixed</h4><h3>The <em>~infallible</em>~ system</h3><p>In November 2020 we installed a 13 kW (AC) with 44 solar panels along with a 20 kWh of Enphase Enpower backup batteries. We live in rural Gold Country California on 10 acres of land. We are on well water which needs electricity for the pump, and when we lose power, we also lose water.</p><p>The allure of the home backup batteries was three fold for us:</p><ol><li>We would never lose power to our well and house as long as the sun was shining.</li><li>We could get the cost of the house batteries, approximately $20,000, reimbursed though the <a href="https://www.cpuc.ca.gov/sgip">Self Generation Incentive Program</a>.</li><li>Compared to buying a propane-fired backup generator which requires expensive-routine maintenance and propane to run, we would have “free” energy along with no maintenance costs.</li></ol><p>Below is a simple photo diagram of the batteries and microgrid system. We can power 100 amps of backup loads, which includes our heat pump HVAC system, refrigeration, stove, many of our main-living area loads, our internet and networking, and most importantly, our well pump.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z6AVjhihUdBHDhi1sE8n8g.png" /><figcaption>The Enphase Encharge batteries, microgrid controller, backup-load circuit breaker (100 amps), and PV combiner in our garage.</figcaption></figure><p>When the grid power goes down, the microgrid controller <em>almost</em> seamlessly switches over to the batteries and solar. We can easily switch over to the microgrid for testing simply by pulling the 200 amp main breaker for the utility (PG&amp;E) meter. During a power outage, we can power our house from both the batteries and solar panels during the day. When the sun goes down, you obviously only have the batteries.</p><h3>Off-grid failure</h3><p>On August 19, 2021, our rural neighborhood suffered a PG&amp;E power failure around 11 pm. Our battery had been drained down to 30%, as it’s scheduled to do to offset our peak-electricity pricing from 4–9 pm. You can see in the Enlighten app below that everything continued to mostly work fine after the outage around 11 pm. And the 100 amps of backup circuits continue to run.</p><p>Apparently the microgrid controller did cause a brownout of sorts when the grid power was cut, as my networking equipment on the backup circuit crashed and our internet was taken offline. This is a new problem, as our internet networking was recently moved to the Encharge backup circuit. My mistake was removing the uninterrupted power supply from the networking equipment; I assumed the power coming off the Enphase microgrid controller was clean enough to prevent the networking equipment from crashing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DCi5lIC6lrMdvoEwD8lqEg.jpeg" /><figcaption>From the Enphase Enlighten Android app at the time of grid failure showing 28% state of charge when the grid went down (look for the “G” symbol). Green is the battery state of charge on the bottom, orange is total consumption, blue is solar generation, grey is grid usage or generation put back to the grid on the top and bottom, respective.</figcaption></figure><h3>Encharge drove itself off a cliff</h3><p>Although the Enphase system appeared to handle the backup situation ok, what happened next is quite disturbing. Around 1 am and about two hours into the power outage I woke up and noticed our air purifier, which is on a backup circuit, was not running. This clued me into the power outage.</p><p>As mentioned previously, the switchover to the microgrid crashed our networking equipment. After a dark trip to the networking dungeon and rebooting everything, I checked the Enlighten app. I found about only 16% of battery charge left. There was no way we were going to make it through the night. We needed to shed some load, ASAP.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HnwrDZd9saE7-uUQ5Izpow.jpeg" /><figcaption>When I woke up at 1 am, the battery was already down to 16% charge (popup model on left side). We were in trouble but the batteries surely will shut themselves off to avoid damage, right?</figcaption></figure><p>At night, we only run our mini-split heat pumps in two of our bedrooms. I immediately shut these heat pumps off from the HVAC app. But the fridge, freezer, and dishwasher were still chipping away at the remaining 15%. I went back to sleep, falsely believing everything would be ok in the morning when the sun started charging the batteries.</p><h4>Encharge killed itself by continually restarting with 3% charge</h4><p>Around 5:15 am, the Encharge batteries reached 3% of remaining charge and switched themselves off for “good”. Things probably would have worked out fine if the batteries actually remained off until the sun was up. But <strong>the microgrid kept trying to restart over itself and over again!</strong></p><p>The Enphase microgrid controller kept trying to come back on, each time reaching a foregone conclusion; the system should shutdown immediately because there was no charge to be had from the solar and the batteries were in a critical state. I know the system was trying to restart because I woke up several times to blinking lights on our HVAC system in our bedroom. Then the HVAC lights would go blank, i.e. the power went off again, and then come back on a bit later. The remaining 3% of battery power, which is needed for the batteries to start charging again from the sun, was squandered trying to restart a system with no power for charging.</p><h4>Morning light</h4><p>When I awoke around 7 am, the lights on each of the six batteries were blinking yellow in “sleep mode”. Eventually the blinking yellow lights on the six batteries began turning off one-by-one until everything was<em> completely dead</em>. Things seemed grim. No grid power, the sun was up and supposed to be charging the batteries, and the ice cream was melting in the freezer. Even though the sky was smokey from wildfire smoke, the system was receiving several kW of solar power and and the microgrid charger still couldn’t restart itself.</p><h4>The well goes dry</h4><p>The worst part of our situation, and the exact reason we’re getting a ~$20,000 <a href="https://www.cpuc.ca.gov/sgip">Self Generation Incentive Program</a> rebate for “Equity Resiliency” for 20 kWh of backup batteries, <em>is to power our well</em>. Our well uses roughly 6–7 amps at 240 volts (1.4 kW) when it’s running.</p><p>The well-pump load is intermittent, thanks to two pressure vessels that “store” the water pressure. But it is a significant-intermittent load. You get one toilet flush for free when the power goes out from the water in the tank. Then you need to find some more water to flush with. We have 90 gallons of emergency water just for this purpose… but the Enphase Encharge batteries were supposed to remove the need to schlep water around the house and pour it into the toilets.</p><h3>Enphase Support</h3><p>I could, and maybe should, write an entire article about Enphase Support. Our solar system installer emailed me on the morning of the power outage to let me know my batteries were critically low on charge. I happened to bicycle to the end of my driveway, where I got cellular-internet connectivity to see an email from my installer about the issue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*adH8tsZ51XOd0N-4OgiJLQ.png" /><figcaption>Errors proactively noticed by my solar installer on the morning of the power outage.</figcaption></figure><p>Our solar installer jumped onto the Enphase chat and <em>waited for over two hours</em> to get a response. The response? We can’t help you. You have to call a special support phone number. When the installer called that number, it was busy and we had to wait for a callback. Meanwhile… the ice cream was still melting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/855/1*2vXIR7wp1_VpJc-K8-RfZg.jpeg" /><figcaption>After waiting on the chat for two hours with our system still dead, our solar system installer was told to call a support number. Once we got through to the phone support, things only got worse.</figcaption></figure><h4>Enphase support did absolutely nothing</h4><p>When we finally got a call back from Enphase, their response basically resembled the sarcastic guy from IT support that asks, “Have you tried rebooting your computer?” By now it was 11 am and we’d been without grid power for 12 hours. But there had been plenty of solar power to charge the house batteries for three hours.</p><p>With my installer on the line, the Enphase tech support had me turn off some of the circuit breakers in the microgrid controller… and simply wait for something to happen. There was absolutely nothing that could be done to bring out dead batteries back except pray the system recovered when the grid came back on.</p><p>Our installer raised the obvious question to the Enphase tech. “Why can’t you configure the batteries to not kill themselves by leaving a bigger cushion of reserve when the batteries power off, currently set at 3%?” our installer asked. This is especially relevant if the batteries cannot be restarted after a complete discharge without the electrical grid, which as we know now, is apparently the case. Enphase had no response or no plan to remedy this issue, according to the technician we spoke to on the phone.</p><h3>Left for dead by Enphase support</h3><p>Enphase has indirectly received millions of California ratepayer dollars via the SGIP incentive. But after this experience, it’s not clear the Encharge batteries are a <em>reliable</em> backup power system. If this was a <em>public safety power shutoff </em>(PSPS), rather than just an unplanned outage, we could have been left for days without power despite having a 13 kW solar system on our roof and 20 kWh of batteries in the garage. Enphase offered no assistance to us in our time of need. We wasted several hours just trying to get a hold of support during a time of limited connectivity.</p><p>Our worst case scenario, in my mind, is now:</p><ol><li>Loosing grid power right after we discharge our home batteries in the evening to offset peak-pricing electricity rates (this happened nearly perfectly during this incident)</li><li>Having limited ability to recharge our battery during the day due to wildfire smoke (this also happened during this incident but the Enphase system was dead… so somewhat moot)</li><li>Having an extended power outage in 100 °F+ temperatures and not being able to cool the house (we got lucky as it was relatively cool when the grid failed this time)</li></ol><p>What is even worse is that some California utility customers are receiving these home-battery backup systems to run medical devices. This is not currently our situation but imagine if we had counted on this backup power to run a critical medical device?</p><h4>Bricked microgrid</h4><p>What I didn’t anticipate was a completely bricked microgrid in our house. A bricked microgrid is completely worthless. You can’t even talk to it to see what’s going on or what is wrong. As Enphase Support said, you just have to see what happens when the grid power turns back on.</p><p>I do sincerely hope that Enphase can come up with a software fix for this pretty obvious problem of system over-depletion. But if Enphase cannot show that its system is reliable, it is questionable whether Enphase systems should continue to be eligible for <a href="https://www.cpuc.ca.gov/sgip">Self Generation Incentive Program</a> funding from California ratepayers. Especially when people are relying on these backup batteries to power their medical devices and water wells.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c71d04753439" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Forecast Verification Using Citizen-Science Rain Gauges]]></title>
            <link>https://medium.com/swlh/forecast-verification-using-citizen-science-rain-gauges-12aaa1d0fac7?source=rss-9713d897e298------2</link>
            <guid isPermaLink="false">https://medium.com/p/12aaa1d0fac7</guid>
            <category><![CDATA[rain]]></category>
            <category><![CDATA[jupyter-notebook]]></category>
            <category><![CDATA[california]]></category>
            <category><![CDATA[weather-forecasts]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Mike Dvorak, PhD]]></dc:creator>
            <pubDate>Tue, 24 Nov 2020 01:25:27 GMT</pubDate>
            <atom:updated>2023-02-16T01:03:09.143Z</atom:updated>
            <content:encoded><![CDATA[<p>Every day, thousands of volunteers faithfully wander out into their backyards to check their <a href="https://cocorahs.org/">CoCoRaHS</a> rain gauges. The beauty of these rain gauges is their simplicity. The CoCoRaHS gauges have no moving parts, they are <a href="https://weatheryourway.com/collections/cocorahs-gauge-parts/products/official-cocorahs-gauge">affordable</a>, more reliable/accurate than <a href="https://blogs.cornell.edu/yourenewa/2015/06/18/accuracy-of-rain-gauge-measurements/">automated rain gauges</a>, and they let you know the <em>exact</em> amount of rain/snow that your yard received. CoCoRaHS rain gauge data is also incredibly useful for the verification of weather forecasts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nc5cht3fm4nvZgoa33IYpg.jpeg" /><figcaption>My CoCoRaHS gauge on my deck in the Oakland Hills, California.</figcaption></figure><p>Being a part of this citizen science network is really gratifying. With CoCoRaHS, you’re helping to create a first-class weather dataset that can be used for both weather and climate research. When I previously lived in Corpus Christi, Texas, I used my CoCoRaHS station as a <a href="https://www.weather.gov/SKYWARN">SKYWARN Storm Spotter</a> to let my local National Weather Service office when we’d received more than 1 inch of rain in an hour, which usually lead to street flooding. Because rain is much more variable than any other meteorological variable, you can get several inches of rain while a mile away a neighborhood may receive little rain.</p><p>Forecasting precipitation is hard, really hard. Because precipitation in weather models is calculated based on many other variables (temperature, humidity, cloud cover, incoming/outgoing radiation, etc…), forecast errors in those variables can really add up to some large forecast errors for precipitation. <em>But forecast errors vary over time… and </em><strong>many forecasts are made for each place and time</strong><em>.</em> So I wanted to know, do precipitation forecasts get more accurate over time? My hypothesis was, yes, forecasts should get better as the <a href="https://w1.weather.gov/glossary/index.php?word=forecast+valid+for">forecast valid time</a> approaches.</p><p>We got plenty of rain this winter in San Francisco Bay Area… about the normal climatological amount. I wanted to see how the <a href="https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/global-forcast-system-gfs">Global Forecast System</a> (GFS) 0.25 degree forecast did in predicting the CoCoRaHS rain amounts for the entire state of California, up to 16 days of time out. Do GFS precipitation forecasts get better over time? Let’s find out.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sW5IxiVMhlKBXMvUmWTbRQ.png" /><figcaption>San Francisco rainfall through April, 2019. We got a lot of rain in February 2019 so I used February 1–8, 2019 as the precipitation forecast validation period. Plot used with permission from <a href="https://www.ggweather.com/water/">Golden Gate Weather Services</a>.</figcaption></figure><p>CoCoRaHS has a somewhat hidden data API. You can find an API request generator <a href="http://data.cocorahs.org/cocorahs/export/exportmanager.aspx">here</a>. All of the CoCoRaHS daily observations from February 1–8, 2019 for California were downloaded. You can see a sample of these these observations in the image below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*00_q057I9I-dCLoIzkKcVQ.png" /><figcaption>California CoCoRaHS observations for February 2, 2019. Used under the Creative Common 3.0 license.</figcaption></figure><p>CoCoRaHS observations have a few unique features. Station observers are encouraged to check their stations as close to 7 am local time as possible. Of course, life gets in the way and it’s not possible for everyone to stick to this schedule. I assumed that any precipitation observation between 5 am and 9 am local was a valid observation, i.e. a 24-hour observation.</p><p>So next, we need a bunch of precip forecasts. Luckily, the National Center for Atmospheric archives the old GFS forecasts <a href="https://rda.ucar.edu/datasets/ds084.1/">here</a> (you need to create a login for free to access the dataset). If you’re more interested in real-time forecasts, you can find these files at the <a href="https://www.nco.ncep.noaa.gov/pmb/products/gfs/">NCEP’s operational forecast site</a>. There are lots of different freely available forecasts, some with much higher spatial resolution like the HRRR at 3-km horizontal resolution. I just chose the 0.25 degree GFS forecast here, roughly 24-km horizontal resolution, to get a general ideal of the precip error for a freely available forecast with global coverage.</p><p>The complete details of the analysis can be found in <a href="https://github.com/wxmiked/cocorahs-verification/blob/master/california-precip-verification-2019-02.ipynb">this Jupyter Notebook</a>. The basic procedure was this:</p><ol><li>Load up a bunch of GFS forecasts into a <a href="https://github.com/wxmiked/windb2">WinDB2</a> Postgres database, an open source geospatial database project I started to store weather and other environmental forecasts.</li><li>Download all of the CoCoRaHS observations from California from February 1–8, 2019. Ignore all observations that fall outside a +/- 2 hour time window from 7 am local time.</li><li>Find the GFS grid point closest to the CoCoRaHS precipitation observation.</li><li>Calculate the error for <em>all</em> forecasts made for a particular CoCoRaHS gauge location and time. Because the GFS forecast is hourly (i.e. every hour) for 0–120 hours, 3-hourly for 120–240 hours, and 12-hourly from 240–384 hours, we use Postgresql database to do some fancy aggregation to make the math easy for us, explained below.</li></ol><h4>Elegant error calculation using SQL</h4><p>Due to the varying forecast intervals talked about in #4 above, calculating which forecast corresponds to which 24-hour precipitation forecast period could be pretty challenging. But since we have everything loaded into Postgres, it was relatively simple to account for this mismatch in forecast interval. The SQL statement from the Jupyter Notebook is shown below. obs represents the <em>observation</em>, fcst is the <em>forecast</em>, and in stands for <em>inches</em>. prateis the GFS forecast variable for total precipitation rate.</p><pre>sql = &quot;&quot;&quot;SELECT <strong>%s</strong> AS station_id, <br>           COUNT(*),<br>           AVG(value)*3600*24*<strong>%s</strong> AS fcst_rain_in_per_day,<br>           AVG(EXTRACT(epoch FROM t - init)/3600) AS fcst_age_hr_avg<br>         FROM prate_1<br>         WHERE geomkey=<strong>%s</strong> AND t&gt;=<strong>%s</strong> AND t&lt;<strong>%s</strong><br>         GROUP BY init <br>         ORDER BY fcst_age_hr_avg;&quot;&quot;&quot;</pre><pre>windb2.curs.execute(sql, <br>                    (obs[&#39;st_num&#39;],<br>                     conv_prate_to_inch_per_sec.magnitude,<br>                     station_meta[&#39;closest_geomkey&#39;], <br>                     obs_t_nominal, <br>                     obs_t_nominal + timedelta(days=1)))</pre><p>If we assume that the forecast amount of rain for the 24-hour CoCoRaHS observation period is simply the average of all forecasts that fall within that period, the calculation is relatively simple. For forecasts 0–120 hours out, you’re always going to have 24-hourly forecasts per day, so the forecast is simply the average precipitation rate for each hour times 24 hours.</p><p>Admittedly, there are some small edge-case errors that are not accounted for here. For example, if your 24-hour forecast includes hours 115 thru 128, the first six forecasts will be hourly and the last eight hours will be three hourly. So technically the forecast hours should be weighted accordingly. But in the interest of simplicity, we ignore this small error in weighting and as we’ll see below, this doesn’t present a significant issue with the results.</p><h4>Do forecasts improve with time?</h4><p>As explained above, generating the precipitation forecast error was relatively easy using the WinDB2 schema and utilities. From the <a href="https://github.com/wxmiked/cocorahs-verification/blob/master/california-precip-verification-2019-02.ipynb">Jupyter notebook</a> using Pandas and Matplotlib to plot histograms of the forecasts and observations (below image), we can immediately see that the GFS forecasts severely under-predict the forecast rain for the study period. Whereas the highest forecast is somewhere around 0.5 inches of rain per day (leftmost plot below), actual observations are as high 1.5 inches per day (rightmost plot below).</p><p>In my experience, the relatively low resolution GFS forecasts under-forecasting rain really isn’t surprising. Much of the heavy rain in California is generate by <a href="https://cloudatlas.wmo.int/orographic-influences-on-the-windward-side.html">orographic effects</a>, i.e. mountains pushing up moist air and generating more rain. At 0.25 degrees horizontal resolution (i.e. about 24x24 km² per model grid cell), you’re not really resolving the sharp coastal and inland mountain ranges that are rain makers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/930/1*ris-JCpyQ_CyqvyursQc2g.png" /><figcaption>Distribution of CoCoRaHS precipitation forecasts (left), age of forecast in hours (middle), and precipitation observations (right)for the 1st week of 2019 in California.</figcaption></figure><p>What is surprising however is that my hypothesis, at least in this case study, proved to be wrong. In fact, precipitation forecast error got worse with time, as can be seen by the blue line below, which represents the root mean square error between the precipitation forecast and the CoCoRaHS observation. Pay careful attention of the x-axis in the plot. The newest forecasts (i.e. the<em> youngest</em> forecast ages) are on the the left-hand side.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/956/1*VAIVN2XjoizNHDnACyJHHA.png" /><figcaption>Precipitation forecast error and descriptive statistics versus the forecast age in hours (all y-axis variables are inches per hour). RMSE is root mean square errors.</figcaption></figure><p>So why didn’t the forecast get more accurate over time? I’m not sure at this point if I have a good answer to that question. This study did however encourage me to create an automated forecast verification website called Weather Verify. You can check the Weather Verify MVP app out <a href="https://app.weathertactics.io/">here</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zQRW0j394g7UnrZHcO7yJg.png" /><figcaption><a href="https://app.weatherverify.com/">Weather Verify MVP app</a> inspired by the forecast verification in this study. The screenshot shows GFS temperature verification in the Mid-Atlantic/NYC region for GFS forecasts between 0–18 hours old, in real time.</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=12aaa1d0fac7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/swlh/forecast-verification-using-citizen-science-rain-gauges-12aaa1d0fac7">Forecast Verification Using Citizen-Science Rain Gauges</a> was originally published in <a href="https://medium.com/swlh">The Startup</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>