주제별 자료/기후

기후 모델의 지역화 (AI시대에도 지리학이 필요한 이유) (Science)

bus333 2025. 6. 15. 07:58

Science - 2025.06.05

 

 

 

NotebookLM이 만든 612초짜리 팟캐스트입니다. (아래를 클릭하면 다운로드 됩니다. 들어보세요~)

Hazy Futures_ Local Impacts of Global Warming.wav
17.06MB

 

 

 

우선, 이 기사를 통해 본 AI시대의 지리학의 중요성을 중심으로 얘기해보겠습니다.

 

오스틴에서 기후 재앙을 예방하는 데 있어 지리학은 중요한 역할을 합니다. 특히, 오스틴 수자원 관리 부서의 '마리사 플로레스 곤잘레스'는 문학 소녀를 꿈꾸고 영문과를 졸업한 뒤 다시 지리학 학위도 받습니다. 지리에 관심이 많았던 그녀는 십 대 때부터 지리 정보 시스템(GIS) 도구를 활용하여 에너지 및 수자원 시설 계획을 돕는 아버지의 사업을 도왔으며, 이후 텍사스 대학교 오스틴 캠퍼스에서 지리학을 전공했습니다.

 

지리학의 중요성은 다음과 같이 나타납니다:

 

인프라 계획 및 준비: GIS 도구는 에너지 및 수자원 시설을 계획하는 데 사용되며, 이는 오스틴이 직면한 인구 증가와 더 더워지는 미래에 대비하는 데 필수적입니다. 오스틴은 현재 인구의 3분의 1 규모일 때 지어졌던 월넛 크릭 폐수 처리장과 같은 대규모 시설을 통해 미래를 대비했습니다.

 

물 공급 관리: 오스틴은 건조한 텍사스 힐 컨트리에 위치하며 항상 텍사스 콜로라도 강에 물을 의존해왔습니다. 과거에는 홍수나 가뭄으로 어려움을 겪었으며, 1950년대 후반에 일련의 댐을 건설하여 저수지를 만들었지만, 2008년부터 2016년까지 최악의 가뭄을 겪었습니다. 이러한 경험은 지리학적 이해를 바탕으로 미래의 가뭄 위험에 대비하고 수자원을 관리하는 계획의 중요성을 부각시킵니다.

 

기후 모델 활용 및 지역적 특성 이해: 마리사 플로레스 곤잘레스와 그녀의 동료들은 가뭄으로 인해 미래에 저수지 수위가 얼마나 더 낮아질지 파악하기 위해 기후 모델을 활용했습니다. 초기 보고서에서는 건조한 날이 더 많아지고, 비가 오는 날에는 더 많이 오는 경향을 예측했지만, 일부 모델은 극심한 폭우로 인해 연간 유량이 지금보다 7배 증가할 수 있다는 예측을 내놓기도 했습니다. 이러한 불확실성은 지역별 기후 모델링의 어려움을 보여주지만, 지리학적 지식을 통해 오스틴의 수자원 계획에 중요한 콜로라도 강 유량 예측으로 전환할 수 있습니다.

 

미래 대비 전략 개발: 새로운 워터 포워드(Water Forward) 보고서는 미래의 물 부족 시나리오에 대비하기 위한 전략들을 포함하고 있습니다. 이에는 처리된 폐수의 재활용 확대, 주민들의 물 사용량 절감 장려, 지하 대수층에 물 저장, 백업 저수지 추가, 지하수 담수화 가능성 연구, 비상 시 폐수 처리장에서 나온 물을 식수로 재사용하는 방안 등이 포함됩니다. 이러한 전략들은 오스틴의 지리적, 수문학적 특성을 고려한 계획입니다.

 

결론적으로, 오스틴은 과거의 가뭄 경험과 미래의 기후 변화 예측을 바탕으로 지리학적 이해와 GIS 도구를 활용하여 도시의 인프라와 수자원 관리 계획을 세우고 있으며, 이는 기후 재앙에 대한 회복탄력성을 높이는 데 필수적인 요소입니다.

 

 

 

---------------------------

 

 

 

기사의 요약 보고서

 

불확실한 미래와 기후 모델링의 도전과제 - 텍사스 오스틴 사례를 중심으로 2025년 6월 5일자 Science지의 메인 기사인 ‘Hazy Futures – Projecting the local impacts of global warming is a stubborn challenge. But cities need answers fast’ 바탕으로 기후 변화의 지역적 영향 예측과 관련한 주요 주제, 가장 중요한 아이디어, 그리고 핵심 사실들을 상세히 검토합니다. 특히 텍사스 오스틴의 사례를 통해 도시들이 직면한 도전과 적응 노력을 조명합니다.

 

 

1. 주요 주제 및 핵심 아이디어

 

이 문서의 핵심 주제는 지구 온난화의 지역적 영향을 예측하는 데 따르는 고질적인 어려움과 불확실성, 그리고 이러한 불확실성 속에서도 도시들이 기후 변화에 적응하기 위해 신속하게 결정을 내려야 할 필요성입니다.

 

지역적 영향 예측의 어려움: 전지구적 기후 모델은 큰 그림에서는 정확하지만, 지역적 규모에서의 온도 및 강수량 예측에는 심각한 불확실성이 존재합니다. "It really is a mess right now," (현재는 정말 엉망이다)라고 캘리포니아 대학교 로스앤젤레스의 기후 과학자 알렉스 홀은 말합니다. 모델들은 유럽의 잦은 폭염이나 동부 열대 태평양의 냉각 등 실제 발생한 지역적 현상들을 예측하지 못했습니다. 시카고 대학교의 기후 역학자 티파니 쇼는 이를 "the other climate crisis" (또 다른 기후 위기)라고 부릅니다.

 

모델 편향 및 한계: 기후 모델은 지구 시스템을 100km 이상의 격자 상자로 나누어 시뮬레이션하기 때문에 구름 형성 등 직접 시뮬레이션할 수 없는 과정에서 경험적 코드 사용으로 인한 편향이 발생합니다. 쇼는 "we were making predictions of worlds we couldn’t see," (우리는 볼 수 없는 세상을 예측하고 있었다)고 언급하며, 이제는 수십 년간의 관측을 통해 모델의 결함을 식별할 수 있게 되었다고 말합니다.

 

다운스케일링 및 편향 보정의 한계: 전지구적 모델의 출력을 지역 모델에 입력하거나 통계적 방법을 사용하는 '다운스케일링 및 편향 보정' 기술이 있지만, 동일한 기후 모델을 사용하더라도 다른 방법은 "drastically different projections" (극적으로 다른 예측)을 초래합니다. 심지어 일부 연구에서는 이러한 조정이 모델이 예측한 추세를 없애버리고, 실제 모델 데이터와 무작위 노이즈의 결과가 통계적으로 구별 불가능할 정도임을 보여주었습니다. 노스이스턴 대학교의 기후 과학자 아우룹 강굴리는 "All the science that goes into building these earth system models was rendered meaningless," (이러한 지구 시스템 모델을 구축하는 데 들어가는 모든 과학이 무의미해졌다)고 지적하며 그 함의가 심각하다고 말합니다.

 

도시의 시급한 대응 필요성: 과학자들이 모델의 한계에 대해 고민하는 동안에도, 도시, 보험 회사, 주택 건설업자, 농부 등은 기후 변화에 대비하기 위한 답을 긴급히 요구하고 있습니다. 인류가 경험하지 못한 미래에 직면하여 기반 시설의 역량과 탄력성에 대한 결정이 도시 단위, 카운티 단위로 신속하게 이루어져야 합니다.

 

상업적 기후 위험 제품 시장의 난립: 모델의 불확실성에도 불구하고, 수많은 기업들이 "unrealistic levels of detail, down to a city block" (도시 블록 단위까지 비현실적인 수준의 세부 정보)을 약속하며 기후 위험 제품을 판매하고 있습니다. 코넬 대학교의 기후 과학자 데이비드 래퍼티는 이를 "a Wild West scenario right now" (지금은 마치 무법천지 같은 상황)라고 묘사하며 제품 평가의 필요성을 강조합니다.

 

 

2. 오스틴, 텍사스의 사례 연구

 

텍사스 오스틴은 기후 변화에 대한 지역적 적응의 중요한 사례를 제공합니다.

 

(오스틴이 왜 위험한지를 다뤘던 포스팅 : 텍사스에서 샌안토니오 다음으로 건물 붕괴 위험이 높음)

https://geowiki.tistory.com/3861

 

미국의 주요 도시들의 침강하는 이유 (WP)

Land under the country’s largest cities is sinking. Here’s where — and why. - The movement is slow — sinking on the scale of millimeters per year in the United States — but the effects accumulate over years. The land underneath the largest cities

geowiki.tistory.com

 

(수돗물 안쓰고 지하수 사용하면 도시 가라 앉는다는 얘기)

https://geowiki.tistory.com/3856

 

가라앉는 미국의 도시들 (NYT)

Across America, Big Cities Are Sinking. Here’s Why. - A major reason is too much groundwater is being pumped out, new research shows, threatening buildings and infrastructure nationwide. A new analysis of America’s 28 largest population centers found t

geowiki.tistory.com

 

 

 

극심한 가뭄 경험: 오스틴은 건조한 텍사스 힐 컨트리에 위치하며 콜로라도 강에 물을 의존해왔습니다. 2008년부터 2016년까지 1950년대 이후 최악의 가뭄을 겪으며 저수지 수위가 2/3로 떨어지고 주민들은 심각한 물 사용 제한에 직면했습니다. 오스틴 워터의 수자원 계획 책임자인 마리사 플로레스 곤잘레스는 "When you’re in the middle of a drought that’s very severe, you don’t know where the bottom is." (심각한 가뭄 한가운데 있을 때는 바닥이 어디인지 알 수 없다)고 말합니다.

 

주 정부와의 정책 차이: 텍사스 주 차원에서는 기후 변화를 미래 가뭄 위험 평가에 반영하지 않지만, 민주당의 거점인 오스틴은 이러한 제약에서 자유롭습니다.

 

"Water Forward" 보고서와 초기 도전: 가뭄에 직면하여 오스틴은 미래 물 공급을 연구하는 "Water Forward" 프로젝트를 시작했습니다. 초기 보고서는 20개 전지구적 기후 모델의 출력을 사용하여, 21세기 후반 콜로라도 강의 연간 유량이 현재보다 30% 감소할 것으로 예측했습니다. 그러나 일부 모델은 극심한 폭우로 인해 연간 유량이 7배 증가할 수 있다는 "Amazonian prediction" (아마존과 같은 예측)을 제시하여, 오스틴이 더 건조해질 뿐만 아니라 더 습해질 준비도 해야 하는지에 대한 의문을 불러일으켰습니다.

 

개선된 모델링 접근법: 두 번째 Water Forward 보고서(2018년 계획 시작)에서는 텍사스 대학교(UT)의 기후 과학자 양종량과 사비하 타바숨이 참여하여 모델링 접근 방식을 개선했습니다.

다양한 배출 시나리오: , , 저 배출 시나리오를 사용하여 더 넓은 가능성을 반영했습니다.

모델 선별 및 평가: 35개 기후 모델을 면밀히 검토하여 텍사스 기후를 가장 잘 포착하는 5개 모델(한국, 호주, 유럽 모델)을 선별했습니다. 양은 "We have high confidence they are giving the right results for the right reasons," (우리는 그 모델들이 올바른 이유로 올바른 결과를 제공한다고 확신한다)고 말합니다.

"모델 평가"의 중요성: 그라츠 대학교의 기후 과학자 더글러스 마론은 "The key aspect is model evaluation." (핵심은 모델 평가)라고 강조하며, 많은 지역 기후 예측이 이 단계를 건너뛴다고 지적합니다.

 

새로운 예측 및 적응 전략: 개선된 모델링 결과는 더 타당했습니다. 건조 일수가 증가하고 비는 더 심한 폭풍우에 집중되었지만, 이전 예측보다 극단적인 유량 증가는 덜 나타났습니다. 인구 증가와 같은 다른 예측과 통합한 결과, 일부 시나리오에서는 2070년에 도시의 물이 고갈되기 시작할 것으로 나타났습니다.

이를 막기 위한 전략으로 폐수 재활용의 대폭 확대("purple pipes"), 주민들의 물 절약 장려, 지하 대수층에 물 저장, 보조 저수지 확보, 지하수 담수화 연구, 그리고 비상시 폐수 처리수 식수 재활용 등이 포함되었습니다. 결론적으로 보고서는 "The future is drierand riskierbut still livable." (미래는 더 건조하고 더 위험하지만, 여전히 살 수 있다)고 말합니다.

 

지속적인 노력: 오스틴은 이미 누수 감지 스마트 수도 계량기를 설치하고, 신축 대형 상업 건물에 물 재활용을 의무화하는 등 적응 노력을 시작했습니다. 플로레스 곤잘레스는 "The next 10 years are going to be some of the most important years in Austin Water’s history." (향후 10년은 오스틴 워터 역사상 가장 중요한 시기가 될 것)이라고 강조합니다.

 

 

3. 미래 전망 및 기술 발전

 

기후 예측의 정확성을 높이기 위한 다양한 노력이 진행 중입니다.

 

다운스케일링 기법 비교 프로젝트: 미국 해양대기청(NOAA)은 마이애미 대학교의 벤자민 커트먼이 공동 주도하는 프로젝트에 자금을 지원하여 다양한 다운스케일링 기법의 성능을 비교하고 평가하고 있습니다. 궁극적인 목표는 지방 정부가 널리 사용할 수 있는 기후 예측 서비스의 기반을 개발하는 것입니다.

 

기후 모델 합성 (Berkeley Earth): 캘리포니아의 기후 비영리단체인 Berkeley Earth는 올해 '기후 모델 합성(Climate Model Synthesis)'이라는 새로운 접근 방식을 선보일 예정입니다. 40개의 전지구적 기후 모델 중 각 지역의 역사적 관측과 가장 잘 일치하는 모델을 찾고, 기계 학습 알고리즘을 사용하여 결과를 다운스케일링할 것입니다. 최고 과학자인 로버트 로드는 이것이 "the breadth of uncertainty and expertise represented by the models" (모델이 나타내는 불확실성과 전문 지식의 폭)을 반영하는 지역적이고 합리적으로 정확한 예측이 될 것이라고 말합니다.

 

초고해상도 전지구 기후 모델: 유럽연합(EU)'데스티네이션 어스(Destination Earth)' 이니셔티브는 3억 유로 이상을 투자하여 5km 해상도의 초고해상도 전지구 모델 프로토타입을 개발했습니다. 이는 "globally consistent map of downscaled information" (전지구적으로 일관된 다운스케일된 정보 지도)를 만드는 것을 목표로 합니다.

 

인공지능(AI)의 활용: 초고해상도 모델의 높은 계산 비용으로 인해 AI를 활용하여 기후 모델을 에뮬레이션하고 실행을 빠르게 반복하는 접근 방식이 주목받고 있습니다. 유럽 중기 예보 센터(ECMWF)와 구글은 AI를 사용하여 고해상도 지역 기후 모델의 다운스케일링을 완전히 대체하는 방안을 연구하고 있습니다. 구글의 연구를 이끈 이그나시오 로페즈-고메즈는 AI"compound extremes such as high heat and winds" (고온과 강풍과 같은 복합적인 극한 현상)을 포착하는 데 특히 뛰어나다고 언급하며, 제한된 계산으로도 그 성능을 널리 활용할 수 있을 것이라고 말합니다.

 

 

4. 결론 및 시사점

 

문서는 전지구적 기후 모델이 큰 그림에서는 성공적이지만, 지역적 영향 예측에는 여전히 심각한 불확실성과 한계가 존재함을 명확히 보여줍니다. 특히 강우량 예측의 경우 다양한 방법론이 극적으로 다른 결과를 초래하며, 현재의 다운스케일링 기법마저도 모델의 과학적 기반을 무의미하게 만들 수 있다는 경고가 나옵니다.

 

그럼에도 불구하고, 오스틴의 사례는 이러한 불확실성 속에서도 도시들이 미래를 계획하고 적응 전략을 수립해야 하는 시급한 필요성을 강조합니다. 오스틴은 모델 평가의 중요성을 인식하고, 더 신뢰할 수 있는 지역 예측을 위해 적극적으로 접근 방식을 개선했습니다. 이는 물 재활용, 절약, 대체 수자원 확보 등 구체적인 적응 계획으로 이어졌습니다.

 

미래에는 더 나은 기후 예측 도구가 등장할 것으로 기대됩니다. 다양한 다운스케일링 기법에 대한 비교 연구, 기계 학습을 활용한 모델 합성, 그리고 궁극적으로는 AI를 통해 고해상도 전지구 모델의 시뮬레이션을 가속화하는 것이 중요한 발전 방향으로 제시됩니다. 이러한 기술 발전은 지역 정부들이 기후 변화의 영향을 더 정확하게 이해하고 효과적인 적응 전략을 수립하는 데 필수적인 정보를 제공할 것입니다. 오스틴의 사례처럼, 현재의 불확실성을 인정하되, 사용 가능한 최선의 정보를 바탕으로 선제적으로 행동하는 것이 중요합니다.

 

 

 

NotebookLM이 제시한 질문을 통해 살펴보는 책의 핵심 내용

 

질문1 : 전 지구적 기후 모델의 지역적 영향을 예측하는 데 있어 주요 난제는 무엇인가요?

 

예측의 불확실성 및 방법론적 혼란: 전 지구적 기후 모델은 전 세계를 조악한 해상도로 시뮬레이션하기 때문에, 이를 계획자들이 원하는 상세한 지역별 온도 및 강우량 예측으로 변환하는 방법에 대한 합의가 거의 없습니다. 동일한 기후 모델을 사용하더라도 서로 다른 방법론은 특히 강우량 측면에서 극명하게 다른 예측을 초래할 수 있습니다. 현재 "정말 엉망진창"이라고 할 정도로 혼란스러운 상황입니다.

 

글로벌 모델의 지역적 편향 및 한계:

글로벌 기후 모델은 전반적인 지구 온난화의 속도와 강도를 정확히 파악함에도 불구하고, 유럽을 괴롭히는 더 빈번한 폭염과 같은 지역적 영향은 종종 예측하지 못했습니다.

연구자들은 예측을 실제 세계에서 발생한 현상과 비교함으로써 모델의 결함을 식별할 수 있게 되었습니다. 이러한 모델의 실패 목록에는 유럽의 폭염, 동부 열대 태평양의 냉각, 북대서양 제트기류의 강화, 그린란드 상공에 정체된 고기압 시스템의 빈번한 발생, 남극해의 냉각 포착 실패, 건조 지역이 더 습해질 것이라는 잘못된 예측 등이 포함됩니다.

이러한 한계로 인해 많은 기후 과학자들은 지역 예측을 회피하고, 극도로 지역적인 예측을 하는 것은 "어리석은 짓"이라고 생각하기도 합니다.

 

계산 제약 및 상세 모델링의 어려움:

대부분의 기후 모델은 계산 제약으로 인해 지구를 100km 이상의 격자 상자로 나누어 조악하게 시뮬레이션합니다.

고해상도로 더 작은 지역을 다루는 모델(다운스케일링 모델)은 폭풍우의 세부 사항이나 산이 날씨에 미치는 영향을 포착할 수 있지만, 여전히 글로벌 모델의 입력을 사용하므로 편향을 상속받아 이를 고려해야 합니다. 또한, 이러한 모델은 계산 비용이 매우 많이 들어서, 글로벌 모델을 실행할 수 있는 슈퍼컴퓨터도 텍사스만을 10km 해상도로 시뮬레이션할 수는 없습니다.

 

다운스케일링 및 편향 보정 방법의 불확실성:

글로벌 모델의 출력을 고해상도로 더 작은 영역을 다루는 다른 모델에 입력하거나, 통계적 방법을 통해 지역적 세부 정보를 추출하는 "다운스케일링 및 편향 보정" 기술이 개발되었지만, 복잡성이 확실성을 가져다주지는 못했습니다.

동일한 기후 모델과 정확히 동일한 지역을 사용하더라도, 이러한 조정은 크게 다른 결과를 초래할 수 있습니다. 예를 들어, 태평양 북서부의 여름 강우량 예측은 방법에 따라 절반으로 줄어들 수 있습니다.

더 충격적인 연구 결과는, 이러한 조정이 기후 모델이 예측한 어떤 경향도 완전히 없앨 수 있다는 것을 보여주었습니다. 실제 기후 모델 데이터와 무작위 노이즈에 동일한 다운스케일링 및 편향 보정 방법을 적용했을 때, 결과가 통계적으로 구별할 수 없었고, 모델에서 얻은 모든 교훈이 사라져 "지구 시스템 모델을 구축하는 데 들어간 모든 과학이 무의미해졌다"고 밝혀졌습니다. 이는 특히 개발도상국에서 실질적인 결과를 초래할 수 있는 심각한 문제입니다.

어떤 다운스케일링 방법이 다른 방법보다 미래 기후를 예측하는 데 더 나은지 결정할 "실질적인 확신이 없습니다".

 

모델 평가의 중요성과 어려움: 지역 기후 예측은 종종 "모델 평가" 단계를 건너뛰는 경우가 많습니다. 연구자들은 편안하게 느끼는 모델을 사용하거나 이미 다운스케일된 글로벌 데이터 세트를 사용하는데, 한 모델이 전 세계 모든 지역을 잘 다루는 경우는 드물기 때문에 문제가 될 수 있습니다. 다운스케일된 정보의 사용자는 종종 강점이나 약점을 알지 못합니다.

 

이러한 난제에도 불구하고, 오스틴은 더 정확한 예측을 위해 텍사스 대학교(UT) 연구진과 협력하여, 다양한 배출 시나리오를 사용하고, 텍사스 기후를 가장 잘 포착하는 기후 모델을 선별하여 예측의 신뢰도를 높이려는 노력을 기울였습니다. 또한, NOAA는 다양한 다운스케일링 기법을 서로 비교하는 프로젝트에 자금을 지원하고 있으며, Berkeley Earth는 역사적 관측치와 가장 잘 맞는 모델을 찾고 기계 학습을 통해 다운스케일하는 새로운 접근 방식을 선보일 예정입니다. EU'Destination Earth'와 같은 이니셔티브는 초고해상도 글로벌 기후 모델을 구축하고 있으며, 인공지능(AI)을 활용하여 모델 실행을 가속화하고 다운스케일링을 대체하려는 연구도 진행 중입니다.

 

 

질문2 : 도시들은 증가하는 기후 불확실성에 직면하여 미래를 어떻게 계획하고 적응하고 있는가요?

 

도시들은 증가하는 기후 불확실성에 직면하여 미래를 계획하고 적응하기 위해 다양한 전략을 사용하고 있습니다. 특히 텍사스 오스틴의 사례를 통해 이러한 노력들을 살펴볼 수 있습니다.

 

1. 장기적인 인프라 계획 및 확장:

 

오스틴의 월넛 크릭 폐수 처리장은 50년 전 도시 규모가 현재의 3분의 1일 때 지어졌지만, 미래를 내다보고 대규모로 건설되어 현재 인구 증가에 대비할 수 있게 했습니다.

 

기후 변화로 인해 더 심각한 가뭄이 예측됨에 따라, 오스틴은 2050년까지 처리된 폐수 재활용량을 10배 이상 늘리는 10억 달러 규모의 처리장 확장 계획을 세우고 있습니다.

 

새로운 대규모 상업용 건물에는 초기 단계부터 물 재활용 계획을 포함하도록 요구됩니다.

 

2. 수자원 관리 및 보존 전략 개발:

 

오스틴은 건조한 텍사스 힐 컨트리에 위치하여 항상 텍사스 콜로라도 강에 물을 의존해왔습니다. 과거 홍수와 가뭄을 겪었으며, 1950년대 후반에 일련의 댐을 건설하여 저수지를 만들었음에도 불구하고 2008년부터 2016년까지 최악의 가뭄을 겪었습니다.

 

이러한 가뭄을 계기로, 오스틴 수자원 관리 부서의 마리사 플로레스 곤잘레스와 동료들은 기후 모델을 활용하여 미래의 더 따뜻한 환경에서 저수지 수위가 얼마나 더 낮아질지 파악하려 했습니다.

 

'워터 포워드(Water Forward)'라는 미래 수자원 공급 연구를 시작했으며, 최신 보고서에는 미래 물 부족 시나리오에 대비하기 위한 여러 전략이 포함되어 있습니다.

재활용수 사용 확대: 도시 특유의 보라색 파이프를 통해 처리된 폐수의 재활용 사용을 대폭 확대합니다.

주민들의 물 사용량 절감 장려: 주민들에게 물 사용량 절감을 장려합니다.

물 저장 및 비상 수원 확보: 가뭄 시 사용하기 위해 전용 지하 대수층에 물을 저장하는 방안을 모색하고, 전력 회사가 사용했던 기존 호수를 백업 저수지로 추가할 예정입니다.

지하수 담수화 연구: 지하수 담수화 가능성을 연구합니다.

비상시 식수 재사용: 비상 시에는 폐수 처리장에서 나온 물을 저수지로 다시 보내 식수로 재사용하는 방안을 준비합니다.

 

누수 감지 스마트 수도 계량기를 거의 모든 주민에게 보급했습니다.

 

다운타운 오스틴의 새 중앙 도서관은 인근 빗물 배수관에서 빗물을 모아 화장실 용수로 사용하는 시스템을 갖추고 있습니다.

 

3. 기후 모델링 전문가와의 협력 및 예측 개선:

 

점점 더 많은 도시, 보험 회사, 건축업자, 농부들이 기후 모델러들에게 답을 구하고 있습니다.

 

오스틴은 텍사스 대학교(UT)의 기후 과학자들과 협력하여 더 나은 예측을 얻으려 노력했습니다.

그들은 고배출, 중간배출, 저배출 시나리오를 모두 사용하여 더 넓은 가능성을 고려했습니다.

존재하는 다양한 편향을 인지하고 모든 기후 모델을 동등하게 유효하다고 보지 않았습니다. 대신 35개의 기후 모델을 선별하여 텍사스 기후를 가장 잘 포착하는 5개의 모델(한국, 호주, 유럽 모델)을 선정했습니다.

모델의 출력을 지역 예측으로 전환한 후, 이 데이터를 오스틴에 중요한 콜로라도 강 유량 예측으로 변환했습니다. 이를 통해 과거 예측에서 나타났던 극단적인 강수량으로 인한 극심한 유량 증가는 덜 나타났고, 건조한 날이 증가하며 비는 더 심한 폭풍우에 집중될 것으로 예측되었습니다.

 

4. 불확실성에도 불구하고 행동:

 

전 지구적 기후 모델은 전반적인 지구 온난화의 속도와 강도를 정확히 파악하지만, 유럽의 폭염과 같은 지역적 영향은 종종 예측하지 못하는 등 지역적 영향을 예측하는 데 여전히 어려움을 겪고 있습니다.

 

모델의 해상도 문제, 계산 제약, 다운스케일링 및 편향 보정 방법의 불확실성 등이 주요 난제입니다. 동일한 모델과 지역을 사용하더라도 조정 방법에 따라 결과가 크게 달라질 수 있습니다. 심지어 일부 연구에서는 이러한 조정이 모델이 예측한 경향을 완전히 없앨 수도 있다고 지적합니다.

 

이러한 어려움에도 불구하고, 오스틴은 기후 예측에 대한 분석에 자부심을 느끼며 충분히 행동할 수 있는 정보를 가지고 있다고 판단하고 있습니다. 플로레스 곤잘레스는 "다음 10년이 오스틴 수자원 역사상 가장 중요한 몇 년이 될 것"이라고 말합니다.

 

미래에는 NOAA의 다운스케일링 기법 비교 프로젝트, Berkeley Earth의 역사적 관측치와 기계 학습을 활용한 새로운 접근 방식, EU'Destination Earth'와 같은 초고해상도 글로벌 기후 모델 구축 이니셔티브, 그리고 AI를 활용하여 모델 실행을 가속화하고 다운스케일링을 대체하려는 연구 등 더 나은 도구들이 개발될 것으로 기대됩니다.

 

요약하자면, 도시들은 기후 불확실성에도 불구하고 인프라 투자, 수자원 관리 전략 개발, 과학적 모델링과의 적극적인 협력을 통해 미래의 기후 재앙에 대비하고 있습니다.

 

 

Hazy futures

- Projecting the local impacts of global warming is a stubborn challenge. But cities need answers fast

 

In 2011, a record drought dried up reservoirs in central Texas, raising fears that global warming will further stress local water supplies. Tony Gutierrez/AP

 

 

Austin, TexasA wastewater treatment plant isn’t top on everyone’s sightseeing itinerary. But on a cloudless, breezy day in April, Agmed Weber, operations manager for the Walnut Creek plantwhich handles more than half of the sewage here in Texas’s booming capital cityis an enthusiastic tour guide. He marvels at the foresight of those who built the jumbo plant a half-century ago, when Austin was a third of its current size. “They were really thinking about the future,” he says. Weber feels a similar pressureto prepare for not only continued population growth, but also a hotter future. Austin is “in a tight space where we need to think ahead,” he says.

 

The plant is dominated by vast outdoor pools in which microbes chew through waste until it is clean enough to return to rivers. But in one corner sit two large tanks and a series of purple pipes that route some of the treated wastewater back to the city to water lawns or flush toilets. Based on a climate study that predicts more severe droughts in Austin’s future, the city is planning, by 2050, to increase the amount of reused wastewater by more than 10 times, as part of a $1 billion expansion of the plant. “That program is going to grow exponentially,” Weber says.

 

Like many cities around the world, Austin is now facing questions about how to build and adapt for a changing climate. A growing number of these citiesas well as insurance companies, home builders, and farmersare turning to climate modelers for answers. But despite decades of effort, forecasting how global warming will play out on a local scale remains a stubborn challenge, riven with uncertainty. There is little agreement on how best to convert climate models, which simulate the entire world at coarse resolutions, into the detailed local forecasts of temperature and rainfall that planners crave. Different methods lead to drastically different projections, especially in terms of rainfalleven when using the same climate model. “It really is a mess right now,” says Alex Hall, a climate scientist at the University of California, Los Angeles.

 

The problem has become more pronounced with the discovery that global climate models, good at the big picture, often miss local impacts that are already painfully evident. For example, despite nailing the overall pace and intensity of global warming, models did not predict the more frequent heat waves that are plaguing Europe. In order to make local forecasts, modelers must not only do the global-to-local conversion, but also adjust for these model biases. Tiffany Shaw, a climate dynamicist at the University of Chicago, calls this “the other climate crisis.” “The cat’s going to get out of the bag, and the question will be, what value does this information have?”

 

 

The Walnut Creek sewage plant in Austin, Texas, plans to boost the reuse of treated water as droughts become more severe and frequent. Austin Water

 

 

While scientists fret about the limitations of their models, others are not as cautious. Dozens of firms have sprung up selling climate risk products, often promising unrealistic levels of detail, down to a city block. “It’s unfortunately closer to a Wild West scenario right now,” says David Lafferty, a climate scientist at Cornell University. “There is a real need to evaluate those products.”

 

But there is little time. City by city, county by county, decisions must be made about the capacity and resilience of infrastructure in a fast-approaching future that humanity has never experienced. Deep in the heart of Texas, reservoirs need to be built and pipes need to be laid.

 

Located in the dry Texas Hill Country, Austin has always depended on the Colorado River for its water. Not that Colorado River, but rather the Texas Colorado River, which slashes nearly 1400 kilometers through the state before emptying into the Gulf of Mexico. The Colorado was a source of life, but also of hardship when it flooded or dried up. Finally, in the late 1950s, the state built a series of dams, creating reservoirs. One sits astride Austin’s downtown: a long placid lake lined with boardwalks, running trails, and the reflections of newly built high-rises.

 

The reservoirs made Austin more resilient but not impregnably so. From 2008 to 2016, the city suffered its worst drought since the 1950s. Reservoir levels dropped by two-thirds. Residents faced severe water-use restrictions, and farmers downstream lost billions of dollars’ worth of crops. It was a scary time, says Marisa Flores Gonzalez, who leads water resources planning at Austin Water, the city’s utility. “When you’re in the middle of a drought that’s very severe, you don’t know where the bottom is.”

 

 

Marisa Flores Gonzalez has led efforts in Austin, Texas, to use climate models to guide the city’s development plans. Austin Water

 

 

Prompted by that drought, Flores Gonzalez and her colleagues wondered how much lower the bottoms might be in a warmer future. “We are a land of extremes,” she says. “We have to be prepared.” At the state level, Texas does not factor climate change into assessments of future drought risk, a policy that may reflect the dominance of climate-skeptical Republicans in its government. But Austin, the state’s leading Democratic stronghold, is under no such restriction.

 

Flores Gonzalez had studied the Texas landscape from an early age. As a teen, she helped her father with his small business, which used geographic information systems (GIS) tools to help energy and water utilities plan infrastructure. In college at the University of Texas (UT) at Austin, she first pursued an English degree. “I was going to be the next great Mexican American writer,” she says. Then the 2008 recession hit, and she switched to geography. “I knew utilities. I knew GIS. And so I adjusted my plans.”

 

She joined Austin Water just as the city kicked off a study of its future water supply, an effort it calls Water Forward. The scientists who contributed to the first report followed standard practice at the time. They took the outputs from 20 different global climate models (using only the most severe greenhouse gas emissions scenarioone now seen as implausible). They took temperature and rain measurements from local weather stations and adjusted them based on the projected climate change signal. Then they used those forecasts to estimate the key factor for water infrastructure: future flows in the Colorado near Austin.

 

 

 

위 그림 해설 (by Grok)

이 그림은 전지구적인 기후 모델(GCM, Global Climate Models)을 지역 수준으로 적용하기 위한 두 가지 주요 방법을 설명합니다. GCM은 지구를 약 100km x 100km 크기의 격자 셀로 나누어 전 세계적인 기후 변화와 warming에 따른 미래를 예측하지만, 도시나 농업 단위처럼 더 작은 지역(약 10km 이하)에서는 세부적인 예측이 어렵습니다. 이를 해결하기 위해 다음과 같은 방법이 사용됩니다:

 

1. 통계적 다운스케일링 (Statistical Downscaling)

• 과거 GCM 결과와 날씨 관측 데이터를 비교하여 모델의 편차를 보정합니다.

• 과거나 현재의 날씨 데이터를 GCM 격자 셀 단위로 평균화해 비교한 후, 미래 예측에 이를 적용합니다.

• 단점:

◦ 과거 날씨와 GCM 간 상관관계가 미래에도 변하지 않는다고 가정하지만, 이는 미래 기후 변화로 달라질 수 있음.

◦ 통계적 방법에 따라 결과가 달라질 수 있음.

◦ 고해상도 예측을 "disaggregate"하는 과정에서 정확도가 떨어질 수 있음.

 

2. 지역 기후 모델 (Regional Climate Models, RCM)

• GCM의 미래 예측 결과를 입력으로 받아 더 작은 격자 셀(예: RCM 미세 격자)로 세분화하여 지역적 예측을 수행합니다.

• 지형, 산악, 호수 등의 세부 지표 데이터를 반영해 더 정밀한 예측이 가능합니다.

• 단점:

◦ GCM의 편향을 그대로 물려받을 수 있음.

◦ 고해상도 계산을 위해 슈퍼컴퓨터가 필요해 일반 사용자에게는 접근이 어려움.

◦ 과거 관측 데이터를 통해 편향 보정이 필요함.

 

이 두 방법 모두 지역 단위로의 예측을 가능하게 하지만, 각각 고유한 한계가 있어 정확한 지역 기후 예측을 위해 추가적인 보완이 필요합니다.

 

 

 

In some ways, the projections that ended up in the 2018 report weren’t surprising: Austin would have more dry days, and on days when it did rain, it would rain more, reflecting the ability of warmer air to hold more water. Overall, annual flows on the Colorado in the last 2 decades of the century would be 30% lower than today’s.

 

But a few outliers among the 20 models suggested that, because of extreme rainstorms, annual flows in some years could end up seven times today’s annual averagean Amazonian prediction that didn’t sit right with Flores Gonzalez and made her question the whole modeling process. Austin was preparing to be drier. But should it really prepare to be wetter, too?

 

When you hear the phrase “climate model,” it’s tempting to think researchers have created a silicon simulacrum of the planet, down to the finest detail. But because of computational constraints, most climate models simulate the planet coarsely, dividing the atmosphere and oceans up into gridlike boxes 100 kilometers wide or more. Using the equations of fluid dynamics, the models calculate how the atmosphere evolves as energy and moisture flow from box to box in discrete time steps.

 

Big picture, the models get a lot right: They’ve precisely charted the nearly 1.5°C rise in temperatures the world has experienced since the 1800s, along with accelerated warming in the Arctic and over the continents. “The basic physics is sound,” says Timothy Palmer, a climate scientist at the University of Oxford. But they do have stubborn biases, particularly when it comes to how winds and moisture flow around the planet. Those biases often emerge not from the equations of fluid dynamics, but rather the empirical codes used to handle processes, such as cloud formation, that the models can’t simulate directly.

 

Until recently, Shaw says, “we were making predictions of worlds we couldn’t see.” But with the warming so strong now, and with several decades of observations in hand, climate scientists can now identify model flaws by comparing their predictions with what has transpired in the real world.

 

Last year, Shaw co-led a workshop that confronted the problem head on, leading to a paper this year in Science Advances that ran through a laundry list of the models’ failures. Not only did they fail to predict heat waves in Europe, but they also missed a cooling in the eastern tropical Pacific Ocean. They overlooked the strengthening of the North Atlantic jet stream in winter, which increases rain and snow in Northern Europe, and they also missed the more frequent occurrence of stalled high-pressure systems over Greenland, which bring sunny days and rapid ice melt. They failed to capture a cooling in the Southern Ocean, and they predicted that arid regions would get moister when they didn’t. “I really think this is a time of reflection and maturation,” Shaw says.

 

 

Although good at projecting global trends, climate models have missed regional phenomena, such as an uptick in clear-sky weather over Greenland that has exacerbated ice melt. Josh Haner/The New York Times via Redux

 

 

Given the models’ limitations, many climate scientists eschew regional projections and think the even harder task of making hyperlocal predictions is a fool’s errand.

 

But a whole field nevertheless arose to do just that. Like investigators in a crime procedural show who magnify a telling detail in a photo, somehow increasing its pixel density in the process, these approaches aim to wring local detail from a coarse global picture.

 

One way to do this “downscaling and bias correction,” as it is known, is by taking the output from a global climate model and feeding it into another model that covers a smaller area at high resolution. Such models can capture the details of storms and the way mountains affect weather, but because they still rely on input from the global models, they inherit biases that must be accounted for. And they are computationally expensive to run: A supercomputer that can run a global model would not have the power to simulate Texas alone at a resolution of 10 kilometers.

 

A more commonand computationally cheapermethod was pioneered several decades ago by Andy Wood, a hydrologist at the National Center for Atmospheric Research. He took daily, local records of temperature and precipitation, averaging them together to match the grid size of a global climate model and its monthly cadence. He could then compare the historical records with the climate model to gauge the magnitude of the model’s bias. Finally, after adjusting the local records based on the model’s forecast for the region, he undid the averaging imposed at the start to create new high-resolution climate projections.

 

By now, these techniques have proliferated into a whole cottage industry of ever more complex statistical matching. But the complexity hasn’t brought certainty. Even when the same climate model is used for the exact same region, the adjustments can produce widely varying results, as Wood and others laid out in a study in Environmental Research Letters (ERL) in April. For example, one method predicts the Pacific Northwest should expect 90 millimeters of summer rain by 2100; under another, that projection drops in half. For other parts of the United States, like the Rocky Mountain West, the choice of method can mean the difference between expecting more rain than the presentor less.

 

-------------

 

Unwelcome surprises

Climate models are good at the big picture of global warming, but at a regional level they have blind spots. With decades of observations in hand, researchers can now identify local climate trends the models failed to predict.

 

(GRAPHIC) V. PENNEY/SCIENCE; (DATA) T. SHAW ET AL., FRONTIERS IN CLIMATE, 6, 1391634 (2024)

 

1 Temperature

Europe has faced unexpected summer heat, whereas the Southern Ocean; eastern, tropical Pacific Ocean; and U.S. Midwest have stayed cooler than predicted.

 

2 Rainfall

Models fail to capture rising rainfall in South America and Australia and drying in East Africa, Europe, and northern India.

 

3. Humidity

Because warmer air can hold more moisture, models predict rising humidity in many arid places. But such changes have not occurred in the U.S. Southwest and elsewhere.

 

4Winds

Unforeseen shifts in wind patterns have led to more stalled, sunny weather over Greenland. They have also strengthened the jet stream over the Atlantic Ocean and intensified storm tracks over the Southern Ocean.

 

-----------------

 

 

The adjustments can also wipe out any trends predicted by a climate model. Last year, Auroop Ganguly, a climate scientist at Northeastern University, and his colleagues applied a widely used downscaling and bias correction method to two different outputs: real climate model data and random noise. To their shock, the results were statistically indistinguishable. Every lesson from the model was lost, they found in a study published in ERL. “All the science that goes into building these earth system models was rendered meaningless,” Ganguly says. Adaptation planners and scientists in developing countries use the technique, so its shortcomings could have practical consequences. “The implications are really serious,” Ganguly says.

 

Wood has grown frustrated as downscaling methods have gotten more elaborate. “We still have no real confidence to determine that one downscaling method is better than another” at projecting future climate, he says. “At some point it becomes unpalatable. It, to some extent, destroys confidence in the results.”

 

He also wonders whether the projections from global models, however coarse and unreliable, are good enough for practical purposes. “It may get you to the same decision for your infrastructure as if you went through that increasingly unpalatable process,” he says.

 

The first Austin Water Forward report was a tremendous learning experience for Flores Gonzalez, who now comfortably drops climate science jargon like “CMIP”the acronym for the Coupled Model Intercomparison Project, of course. When city officials began planning a second report in 2018, they wanted the best possible projections. They created an advisory group to provide real-time peer review of the process. And this time the city hired climate scientists at UT, including Zong-Liang Yang and his graduate student at the time, Sabiha Tabassum.

 

Yang and Tabassum made several changes at the start. First, they used projections for high-, medium-, and low-emission futures, representing a broader spread in possibility. Second, knowing the different biases that existed, they didn’t treat every climate model as equally valid. Rather, they sifted through 35 climate models and found the ones that best captured the Texas climate. This filtering led them to five models, from South Korea, Australia, and Europe. “We have high confidence they are giving the right results for the right reasons,” Yang says.

 

All too often, local climate projections skip this step, says Douglas Maraun, a climate scientist at the University of Graz. “The key aspect is model evaluation.” Often researchers use the model they are most comfortable with, whatever it saysor they use a global suite of data that has already been downscaled. Such global data sets are particularly problematic, as rarely does one model handle everywhere in the world well, says John Nielsen-Gammon, the state’s climatologist at Texas A&M University. “The users of downscaled information typically do not know what the strengths or weaknesses are.”

 

 

The fitful flows of the Texas Colorado River have long been a source of life and destruction for Austin, Texas.Jordan Vonderhaar/Bloomberg via Getty Images

 

 

The UT team converted the models’ output into a regional forecast using standard downscaling and bias correction methods. But because the models tracked the region so well, Yang says, the corrections were relatively minor. After that, the data were converted into Colorado River flow, the numbers that mattered for Austin.

 

The results looked more plausible than before. The number of dry days increased, while rain grew concentrated into more severe storms. “But those extreme flows were less prevalent in this round than the last one,” Tabassum says. After integrating these projections with others, such as population growth, the team found that, in half of the scenarios considered, the city would begin to run out of water in 2070.

 

The new Water Forward report, approved by the Austin city council late last year, includes strategies to forestall that future. The city should drastically expand its use of reclaimed water in its distinctive purple pipes. It should encourage its residents to use less water. It should explore storing water in a dedicated underground aquifer for use during drought; it should add a backup reservoir, an existing lake previously used by a power utility; and it should study the possibility of desalinating groundwater. Finally, the report said, the city should prepare for true emergency scenarios where treated water from its sewage plant would be fed back into a reservoir and reused as drinking water.

 

In short, it said: The future is drierand riskierbut still livable.

 

The next time Austinor any other citysets out to forecast its climate future, it may have better tools. Last year, the National Oceanic and Atmospheric Administration funded a project co-led by Benjamin Kirtman, an atmospheric scientist at the University of Miami, to pit different downscaling techniques against each other, applying them to different climate models and seeing how the results diverge. “Do this over and over again and you can ultimately assess how well they work,” he says. His ultimate aim is to help develop the underpinnings of a climate projection service for wide use by local governments.

 

Later this year, Berkeley Earth, a California climate nonprofit previously known for its focus on historical climate records, will debut another approach, which it calls its Climate Model Synthesis. The project will take 40 global climate models and, for each region in the world, find the ones that best match historical observations. It will then use a machine-learning algorithm informed by Berkeley Earth’s high-resolution temperature record to downscale the results. The end result should be local, reasonably accurate projections that reflect “the breadth of uncertainty and expertise represented by the models,” says Robert Rohde, Berkeley Earth’s chief scientist.

 

Another route is to forgo downscaling and bias correction and instead build global climate models at superhigh resolutions. That’s the drive at Destination Earth, an EU initiative costing more than 300 million that has now run prototype global models at a 5-kilometer resolution, using some of the world’s most powerful supercomputers. “The idea is to operationalize climate projections for the first time” by regularly updating these projects, says Irina Sandu, its director. “It’s creating a globally consistent map of downscaled information.”

 

But even with all its computing power, the team has only been able to complete projections out to 2040. And running the models multiple times to capture the spread of possibilities is prohibitively expensive. That’s why Sandu and others believe they will need to take advantage of the promising ability of artificial intelligence (AI) to emulate climate models and repeat their runs quicklyan approach that is already catching on for ordinary weather forecasting, with the European Centre for Medium-Range Weather Forecasts (ECMWF), the world’s leading forecaster, pioneering it.

 

ECMWF and Google, among others, have teams looking at how to use such advanced AI to entirely replace the downscaling done by high-resolution regional climate models. After training on a climate model, Google’s AI substitute proved particularly good at capturing compound extremes such as high heat and winds. “It allows us to really sample all the possible states of the atmosphere,” says Ignacio Lopez-Gomez, who led a Google study published in April in the Proceedings of the National Academy of Sciences. And the limited computation needed to run it could make its power widely available.

 

Early one morning back in downtown Austin, Boyce Freitag descended into an underground concrete cistern the size of a large rowhouse. From a balcony, he looked onto the dark pool below. Light from a hole in the ceiling illuminated the grimy patina of the cistern’s walls. “Hope spiders don’t give you the skeevies,” he says. Originally built for a defunct power plant, the cistern found a purpose again last decade when Austin built its new central library next door. “This is where the magic happens,” adds Freitag, the library’s building manager.

 

The cistern collects rainwater from neighborhood storm drains, which is then pumped to the gleaming library to flush toilets. It’s just one example of where the city is going, Flores Gonzalez says. After the first Water Forward report, the city rolled out leak-detecting smart water meters to nearly every resident. And now new, large commercial buildings will be required to incorporate water reuse into their plans from the start.

 

Austin Water’s collaboration with the UT researchers will continue. Flores Gonzalez takes pride in their analysis of Austin’s future, but they still have more to do: They want to better understand just how much river flow could increase in future years when a warmer atmosphere delivers exceptional rainfall, for example. But Austin knows enough to act, she says. “The next 10 years are going to be some of the most important years in Austin Water’s history.”

 

The adaptation can’t come soon enough. Several years ago, the city once again plunged into drought. And its bottom is still unknown.

 

doi: 10.1126/science.zmq4rob